113 skills found · Page 1 of 4
sayantann11 / All Classification Templetes For MLClassification - Machine Learning This is ‘Classification’ tutorial which is a part of the Machine Learning course offered by Simplilearn. We will learn Classification algorithms, types of classification algorithms, support vector machines(SVM), Naive Bayes, Decision Tree and Random Forest Classifier in this tutorial. Objectives Let us look at some of the objectives covered under this section of Machine Learning tutorial. Define Classification and list its algorithms Describe Logistic Regression and Sigmoid Probability Explain K-Nearest Neighbors and KNN classification Understand Support Vector Machines, Polynomial Kernel, and Kernel Trick Analyze Kernel Support Vector Machines with an example Implement the Naïve Bayes Classifier Demonstrate Decision Tree Classifier Describe Random Forest Classifier Classification: Meaning Classification is a type of supervised learning. It specifies the class to which data elements belong to and is best used when the output has finite and discrete values. It predicts a class for an input variable as well. There are 2 types of Classification: Binomial Multi-Class Classification: Use Cases Some of the key areas where classification cases are being used: To find whether an email received is a spam or ham To identify customer segments To find if a bank loan is granted To identify if a kid will pass or fail in an examination Classification: Example Social media sentiment analysis has two potential outcomes, positive or negative, as displayed by the chart given below. https://www.simplilearn.com/ice9/free_resources_article_thumb/classification-example-machine-learning.JPG This chart shows the classification of the Iris flower dataset into its three sub-species indicated by codes 0, 1, and 2. https://www.simplilearn.com/ice9/free_resources_article_thumb/iris-flower-dataset-graph.JPG The test set dots represent the assignment of new test data points to one class or the other based on the trained classifier model. Types of Classification Algorithms Let’s have a quick look into the types of Classification Algorithm below. Linear Models Logistic Regression Support Vector Machines Nonlinear models K-nearest Neighbors (KNN) Kernel Support Vector Machines (SVM) Naïve Bayes Decision Tree Classification Random Forest Classification Logistic Regression: Meaning Let us understand the Logistic Regression model below. This refers to a regression model that is used for classification. This method is widely used for binary classification problems. It can also be extended to multi-class classification problems. Here, the dependent variable is categorical: y ϵ {0, 1} A binary dependent variable can have only two values, like 0 or 1, win or lose, pass or fail, healthy or sick, etc In this case, you model the probability distribution of output y as 1 or 0. This is called the sigmoid probability (σ). If σ(θ Tx) > 0.5, set y = 1, else set y = 0 Unlike Linear Regression (and its Normal Equation solution), there is no closed form solution for finding optimal weights of Logistic Regression. Instead, you must solve this with maximum likelihood estimation (a probability model to detect the maximum likelihood of something happening). It can be used to calculate the probability of a given outcome in a binary model, like the probability of being classified as sick or passing an exam. https://www.simplilearn.com/ice9/free_resources_article_thumb/logistic-regression-example-graph.JPG Sigmoid Probability The probability in the logistic regression is often represented by the Sigmoid function (also called the logistic function or the S-curve): https://www.simplilearn.com/ice9/free_resources_article_thumb/sigmoid-function-machine-learning.JPG In this equation, t represents data values * the number of hours studied and S(t) represents the probability of passing the exam. Assume sigmoid function: https://www.simplilearn.com/ice9/free_resources_article_thumb/sigmoid-probability-machine-learning.JPG g(z) tends toward 1 as z -> infinity , and g(z) tends toward 0 as z -> infinity K-nearest Neighbors (KNN) K-nearest Neighbors algorithm is used to assign a data point to clusters based on similarity measurement. It uses a supervised method for classification. The steps to writing a k-means algorithm are as given below: https://www.simplilearn.com/ice9/free_resources_article_thumb/knn-distribution-graph-machine-learning.JPG Choose the number of k and a distance metric. (k = 5 is common) Find k-nearest neighbors of the sample that you want to classify Assign the class label by majority vote. KNN Classification A new input point is classified in the category such that it has the most number of neighbors from that category. For example: https://www.simplilearn.com/ice9/free_resources_article_thumb/knn-classification-machine-learning.JPG Classify a patient as high risk or low risk. Mark email as spam or ham. Keen on learning about Classification Algorithms in Machine Learning? Click here! Support Vector Machine (SVM) Let us understand Support Vector Machine (SVM) in detail below. SVMs are classification algorithms used to assign data to various classes. They involve detecting hyperplanes which segregate data into classes. SVMs are very versatile and are also capable of performing linear or nonlinear classification, regression, and outlier detection. Once ideal hyperplanes are discovered, new data points can be easily classified. https://www.simplilearn.com/ice9/free_resources_article_thumb/support-vector-machines-graph-machine-learning.JPG The optimization objective is to find “maximum margin hyperplane” that is farthest from the closest points in the two classes (these points are called support vectors). In the given figure, the middle line represents the hyperplane. SVM Example Let’s look at this image below and have an idea about SVM in general. Hyperplanes with larger margins have lower generalization error. The positive and negative hyperplanes are represented by: https://www.simplilearn.com/ice9/free_resources_article_thumb/positive-negative-hyperplanes-machine-learning.JPG Classification of any new input sample xtest : If w0 + wTxtest > 1, the sample xtest is said to be in the class toward the right of the positive hyperplane. If w0 + wTxtest < -1, the sample xtest is said to be in the class toward the left of the negative hyperplane. When you subtract the two equations, you get: https://www.simplilearn.com/ice9/free_resources_article_thumb/equation-subtraction-machine-learning.JPG Length of vector w is (L2 norm length): https://www.simplilearn.com/ice9/free_resources_article_thumb/length-of-vector-machine-learning.JPG You normalize with the length of w to arrive at: https://www.simplilearn.com/ice9/free_resources_article_thumb/normalize-equation-machine-learning.JPG SVM: Hard Margin Classification Given below are some points to understand Hard Margin Classification. The left side of equation SVM-1 given above can be interpreted as the distance between the positive (+ve) and negative (-ve) hyperplanes; in other words, it is the margin that can be maximized. Hence the objective of the function is to maximize with the constraint that the samples are classified correctly, which is represented as : https://www.simplilearn.com/ice9/free_resources_article_thumb/hard-margin-classification-machine-learning.JPG This means that you are minimizing ‖w‖. This also means that all positive samples are on one side of the positive hyperplane and all negative samples are on the other side of the negative hyperplane. This can be written concisely as : https://www.simplilearn.com/ice9/free_resources_article_thumb/hard-margin-classification-formula.JPG Minimizing ‖w‖ is the same as minimizing. This figure is better as it is differentiable even at w = 0. The approach listed above is called “hard margin linear SVM classifier.” SVM: Soft Margin Classification Given below are some points to understand Soft Margin Classification. To allow for linear constraints to be relaxed for nonlinearly separable data, a slack variable is introduced. (i) measures how much ith instance is allowed to violate the margin. The slack variable is simply added to the linear constraints. https://www.simplilearn.com/ice9/free_resources_article_thumb/soft-margin-calculation-machine-learning.JPG Subject to the above constraints, the new objective to be minimized becomes: https://www.simplilearn.com/ice9/free_resources_article_thumb/soft-margin-calculation-formula.JPG You have two conflicting objectives now—minimizing slack variable to reduce margin violations and minimizing to increase the margin. The hyperparameter C allows us to define this trade-off. Large values of C correspond to larger error penalties (so smaller margins), whereas smaller values of C allow for higher misclassification errors and larger margins. https://www.simplilearn.com/ice9/free_resources_article_thumb/machine-learning-certification-video-preview.jpg SVM: Regularization The concept of C is the reverse of regularization. Higher C means lower regularization, which increases bias and lowers the variance (causing overfitting). https://www.simplilearn.com/ice9/free_resources_article_thumb/concept-of-c-graph-machine-learning.JPG IRIS Data Set The Iris dataset contains measurements of 150 IRIS flowers from three different species: Setosa Versicolor Viriginica Each row represents one sample. Flower measurements in centimeters are stored as columns. These are called features. IRIS Data Set: SVM Let’s train an SVM model using sci-kit-learn for the Iris dataset: https://www.simplilearn.com/ice9/free_resources_article_thumb/svm-model-graph-machine-learning.JPG Nonlinear SVM Classification There are two ways to solve nonlinear SVMs: by adding polynomial features by adding similarity features Polynomial features can be added to datasets; in some cases, this can create a linearly separable dataset. https://www.simplilearn.com/ice9/free_resources_article_thumb/nonlinear-classification-svm-machine-learning.JPG In the figure on the left, there is only 1 feature x1. This dataset is not linearly separable. If you add x2 = (x1)2 (figure on the right), the data becomes linearly separable. Polynomial Kernel In sci-kit-learn, one can use a Pipeline class for creating polynomial features. Classification results for the Moons dataset are shown in the figure. https://www.simplilearn.com/ice9/free_resources_article_thumb/polynomial-kernel-machine-learning.JPG Polynomial Kernel with Kernel Trick Let us look at the image below and understand Kernel Trick in detail. https://www.simplilearn.com/ice9/free_resources_article_thumb/polynomial-kernel-with-kernel-trick.JPG For large dimensional datasets, adding too many polynomial features can slow down the model. You can apply a kernel trick with the effect of polynomial features without actually adding them. The code is shown (SVC class) below trains an SVM classifier using a 3rd-degree polynomial kernel but with a kernel trick. https://www.simplilearn.com/ice9/free_resources_article_thumb/polynomial-kernel-equation-machine-learning.JPG The hyperparameter coefθ controls the influence of high-degree polynomials. Kernel SVM Let us understand in detail about Kernel SVM. Kernel SVMs are used for classification of nonlinear data. In the chart, nonlinear data is projected into a higher dimensional space via a mapping function where it becomes linearly separable. https://www.simplilearn.com/ice9/free_resources_article_thumb/kernel-svm-machine-learning.JPG In the higher dimension, a linear separating hyperplane can be derived and used for classification. A reverse projection of the higher dimension back to original feature space takes it back to nonlinear shape. As mentioned previously, SVMs can be kernelized to solve nonlinear classification problems. You can create a sample dataset for XOR gate (nonlinear problem) from NumPy. 100 samples will be assigned the class sample 1, and 100 samples will be assigned the class label -1. https://www.simplilearn.com/ice9/free_resources_article_thumb/kernel-svm-graph-machine-learning.JPG As you can see, this data is not linearly separable. https://www.simplilearn.com/ice9/free_resources_article_thumb/kernel-svm-non-separable.JPG You now use the kernel trick to classify XOR dataset created earlier. https://www.simplilearn.com/ice9/free_resources_article_thumb/kernel-svm-xor-machine-learning.JPG Naïve Bayes Classifier What is Naive Bayes Classifier? Have you ever wondered how your mail provider implements spam filtering or how online news channels perform news text classification or even how companies perform sentiment analysis of their audience on social media? All of this and more are done through a machine learning algorithm called Naive Bayes Classifier. Naive Bayes Named after Thomas Bayes from the 1700s who first coined this in the Western literature. Naive Bayes classifier works on the principle of conditional probability as given by the Bayes theorem. Advantages of Naive Bayes Classifier Listed below are six benefits of Naive Bayes Classifier. Very simple and easy to implement Needs less training data Handles both continuous and discrete data Highly scalable with the number of predictors and data points As it is fast, it can be used in real-time predictions Not sensitive to irrelevant features Bayes Theorem We will understand Bayes Theorem in detail from the points mentioned below. According to the Bayes model, the conditional probability P(Y|X) can be calculated as: P(Y|X) = P(X|Y)P(Y) / P(X) This means you have to estimate a very large number of P(X|Y) probabilities for a relatively small vector space X. For example, for a Boolean Y and 30 possible Boolean attributes in the X vector, you will have to estimate 3 billion probabilities P(X|Y). To make it practical, a Naïve Bayes classifier is used, which assumes conditional independence of P(X) to each other, with a given value of Y. This reduces the number of probability estimates to 2*30=60 in the above example. Naïve Bayes Classifier for SMS Spam Detection Consider a labeled SMS database having 5574 messages. It has messages as given below: https://www.simplilearn.com/ice9/free_resources_article_thumb/naive-bayes-spam-machine-learning.JPG Each message is marked as spam or ham in the data set. Let’s train a model with Naïve Bayes algorithm to detect spam from ham. The message lengths and their frequency (in the training dataset) are as shown below: https://www.simplilearn.com/ice9/free_resources_article_thumb/naive-bayes-spam-spam-detection.JPG Analyze the logic you use to train an algorithm to detect spam: Split each message into individual words/tokens (bag of words). Lemmatize the data (each word takes its base form, like “walking” or “walked” is replaced with “walk”). Convert data to vectors using scikit-learn module CountVectorizer. Run TFIDF to remove common words like “is,” “are,” “and.” Now apply scikit-learn module for Naïve Bayes MultinomialNB to get the Spam Detector. This spam detector can then be used to classify a random new message as spam or ham. Next, the accuracy of the spam detector is checked using the Confusion Matrix. For the SMS spam example above, the confusion matrix is shown on the right. Accuracy Rate = Correct / Total = (4827 + 592)/5574 = 97.21% Error Rate = Wrong / Total = (155 + 0)/5574 = 2.78% https://www.simplilearn.com/ice9/free_resources_article_thumb/confusion-matrix-machine-learning.JPG Although confusion Matrix is useful, some more precise metrics are provided by Precision and Recall. https://www.simplilearn.com/ice9/free_resources_article_thumb/precision-recall-matrix-machine-learning.JPG Precision refers to the accuracy of positive predictions. https://www.simplilearn.com/ice9/free_resources_article_thumb/precision-formula-machine-learning.JPG Recall refers to the ratio of positive instances that are correctly detected by the classifier (also known as True positive rate or TPR). https://www.simplilearn.com/ice9/free_resources_article_thumb/recall-formula-machine-learning.JPG Precision/Recall Trade-off To detect age-appropriate videos for kids, you need high precision (low recall) to ensure that only safe videos make the cut (even though a few safe videos may be left out). The high recall is needed (low precision is acceptable) in-store surveillance to catch shoplifters; a few false alarms are acceptable, but all shoplifters must be caught. Learn about Naive Bayes in detail. Click here! Decision Tree Classifier Some aspects of the Decision Tree Classifier mentioned below are. Decision Trees (DT) can be used both for classification and regression. The advantage of decision trees is that they require very little data preparation. They do not require feature scaling or centering at all. They are also the fundamental components of Random Forests, one of the most powerful ML algorithms. Unlike Random Forests and Neural Networks (which do black-box modeling), Decision Trees are white box models, which means that inner workings of these models are clearly understood. In the case of classification, the data is segregated based on a series of questions. Any new data point is assigned to the selected leaf node. https://www.simplilearn.com/ice9/free_resources_article_thumb/decision-tree-classifier-machine-learning.JPG Start at the tree root and split the data on the feature using the decision algorithm, resulting in the largest information gain (IG). This splitting procedure is then repeated in an iterative process at each child node until the leaves are pure. This means that the samples at each node belonging to the same class. In practice, you can set a limit on the depth of the tree to prevent overfitting. The purity is compromised here as the final leaves may still have some impurity. The figure shows the classification of the Iris dataset. https://www.simplilearn.com/ice9/free_resources_article_thumb/decision-tree-classifier-graph.JPG IRIS Decision Tree Let’s build a Decision Tree using scikit-learn for the Iris flower dataset and also visualize it using export_graphviz API. https://www.simplilearn.com/ice9/free_resources_article_thumb/iris-decision-tree-machine-learning.JPG The output of export_graphviz can be converted into png format: https://www.simplilearn.com/ice9/free_resources_article_thumb/iris-decision-tree-output.JPG Sample attribute stands for the number of training instances the node applies to. Value attribute stands for the number of training instances of each class the node applies to. Gini impurity measures the node’s impurity. A node is “pure” (gini=0) if all training instances it applies to belong to the same class. https://www.simplilearn.com/ice9/free_resources_article_thumb/impurity-formula-machine-learning.JPG For example, for Versicolor (green color node), the Gini is 1-(0/54)2 -(49/54)2 -(5/54) 2 ≈ 0.168 https://www.simplilearn.com/ice9/free_resources_article_thumb/iris-decision-tree-sample.JPG Decision Boundaries Let us learn to create decision boundaries below. For the first node (depth 0), the solid line splits the data (Iris-Setosa on left). Gini is 0 for Setosa node, so no further split is possible. The second node (depth 1) splits the data into Versicolor and Virginica. If max_depth were set as 3, a third split would happen (vertical dotted line). https://www.simplilearn.com/ice9/free_resources_article_thumb/decision-tree-boundaries.JPG For a sample with petal length 5 cm and petal width 1.5 cm, the tree traverses to depth 2 left node, so the probability predictions for this sample are 0% for Iris-Setosa (0/54), 90.7% for Iris-Versicolor (49/54), and 9.3% for Iris-Virginica (5/54) CART Training Algorithm Scikit-learn uses Classification and Regression Trees (CART) algorithm to train Decision Trees. CART algorithm: Split the data into two subsets using a single feature k and threshold tk (example, petal length < “2.45 cm”). This is done recursively for each node. k and tk are chosen such that they produce the purest subsets (weighted by their size). The objective is to minimize the cost function as given below: https://www.simplilearn.com/ice9/free_resources_article_thumb/cart-training-algorithm-machine-learning.JPG The algorithm stops executing if one of the following situations occurs: max_depth is reached No further splits are found for each node Other hyperparameters may be used to stop the tree: min_samples_split min_samples_leaf min_weight_fraction_leaf max_leaf_nodes Gini Impurity or Entropy Entropy is one more measure of impurity and can be used in place of Gini. https://www.simplilearn.com/ice9/free_resources_article_thumb/gini-impurity-entrophy.JPG It is a degree of uncertainty, and Information Gain is the reduction that occurs in entropy as one traverses down the tree. Entropy is zero for a DT node when the node contains instances of only one class. Entropy for depth 2 left node in the example given above is: https://www.simplilearn.com/ice9/free_resources_article_thumb/entrophy-for-depth-2.JPG Gini and Entropy both lead to similar trees. DT: Regularization The following figure shows two decision trees on the moons dataset. https://www.simplilearn.com/ice9/free_resources_article_thumb/dt-regularization-machine-learning.JPG The decision tree on the right is restricted by min_samples_leaf = 4. The model on the left is overfitting, while the model on the right generalizes better. Random Forest Classifier Let us have an understanding of Random Forest Classifier below. A random forest can be considered an ensemble of decision trees (Ensemble learning). Random Forest algorithm: Draw a random bootstrap sample of size n (randomly choose n samples from the training set). Grow a decision tree from the bootstrap sample. At each node, randomly select d features. Split the node using the feature that provides the best split according to the objective function, for instance by maximizing the information gain. Repeat the steps 1 to 2 k times. (k is the number of trees you want to create, using a subset of samples) Aggregate the prediction by each tree for a new data point to assign the class label by majority vote (pick the group selected by the most number of trees and assign new data point to that group). Random Forests are opaque, which means it is difficult to visualize their inner workings. https://www.simplilearn.com/ice9/free_resources_article_thumb/random-forest-classifier-graph.JPG However, the advantages outweigh their limitations since you do not have to worry about hyperparameters except k, which stands for the number of decision trees to be created from a subset of samples. RF is quite robust to noise from the individual decision trees. Hence, you need not prune individual decision trees. The larger the number of decision trees, the more accurate the Random Forest prediction is. (This, however, comes with higher computation cost). Key Takeaways Let us quickly run through what we have learned so far in this Classification tutorial. Classification algorithms are supervised learning methods to split data into classes. They can work on Linear Data as well as Nonlinear Data. Logistic Regression can classify data based on weighted parameters and sigmoid conversion to calculate the probability of classes. K-nearest Neighbors (KNN) algorithm uses similar features to classify data. Support Vector Machines (SVMs) classify data by detecting the maximum margin hyperplane between data classes. Naïve Bayes, a simplified Bayes Model, can help classify data using conditional probability models. Decision Trees are powerful classifiers and use tree splitting logic until pure or somewhat pure leaf node classes are attained. Random Forests apply Ensemble Learning to Decision Trees for more accurate classification predictions. Conclusion This completes ‘Classification’ tutorial. In the next tutorial, we will learn 'Unsupervised Learning with Clustering.'
chrisneagu / FTC Skystone Dark Angels Romania 2020NOTICE This repository contains the public FTC SDK for the SKYSTONE (2019-2020) competition season. If you are looking for the current season's FTC SDK software, please visit the new and permanent home of the public FTC SDK: FtcRobotController repository Welcome! This GitHub repository contains the source code that is used to build an Android app to control a FIRST Tech Challenge competition robot. To use this SDK, download/clone the entire project to your local computer. Getting Started If you are new to robotics or new to the FIRST Tech Challenge, then you should consider reviewing the FTC Blocks Tutorial to get familiar with how to use the control system: FTC Blocks Online Tutorial Even if you are an advanced Java programmer, it is helpful to start with the FTC Blocks tutorial, and then migrate to the OnBot Java Tool or to Android Studio afterwards. Downloading the Project If you are an Android Studio programmer, there are several ways to download this repo. Note that if you use the Blocks or OnBot Java Tool to program your robot, then you do not need to download this repository. If you are a git user, you can clone the most current version of the repository: git clone https://github.com/FIRST-Tech-Challenge/SKYSTONE.git Or, if you prefer, you can use the "Download Zip" button available through the main repository page. Downloading the project as a .ZIP file will keep the size of the download manageable. You can also download the project folder (as a .zip or .tar.gz archive file) from the Downloads subsection of the Releases page for this repository. Once you have downloaded and uncompressed (if needed) your folder, you can use Android Studio to import the folder ("Import project (Eclipse ADT, Gradle, etc.)"). Getting Help User Documentation and Tutorials FIRST maintains online documentation with information and tutorials on how to use the FIRST Tech Challenge software and robot control system. You can access this documentation using the following link: SKYSTONE Online Documentation Note that the online documentation is an "evergreen" document that is constantly being updated and edited. It contains the most current information about the FIRST Tech Challenge software and control system. Javadoc Reference Material The Javadoc reference documentation for the FTC SDK is now available online. Click on the following link to view the FTC SDK Javadoc documentation as a live website: FTC Javadoc Documentation Documentation for the FTC SDK is also included with this repository. There is a subfolder called "doc" which contains several subfolders: The folder "apk" contains the .apk files for the FTC Driver Station and FTC Robot Controller apps. The folder "javadoc" contains the JavaDoc user documentation for the FTC SDK. Online User Forum For technical questions regarding the Control System or the FTC SDK, please visit the FTC Technology forum: FTC Technology Forum Release Information Version 5.5 (20200824-090813) Version 5.5 requires Android Studio 4.0 or later. New features Adds support for calling custom Java classes from Blocks OpModes (fixes SkyStone issue #161). Classes must be in the org.firstinspires.ftc.teamcode package. Methods must be public static and have no more than 21 parameters. Parameters declared as OpMode, LinearOpMode, Telemetry, and HardwareMap are supported and the argument is provided automatically, regardless of the order of the parameters. On the block, the sockets for those parameters are automatically filled in. Parameters declared as char or java.lang.Character will accept any block that returns text and will only use the first character in the text. Parameters declared as boolean or java.lang.Boolean will accept any block that returns boolean. Parameters declared as byte, java.lang.Byte, short, java.lang.Short, int, java.lang.Integer, long, or java.lang.Long, will accept any block that returns a number and will round that value to the nearest whole number. Parameters declared as float, java.lang.Float, double, java.lang.Double will accept any block that returns a number. Adds telemetry API method for setting display format Classic Monospace HTML (certain tags only) Adds blocks support for switching cameras. Adds Blocks support for TensorFlow Object Detection with a custom model. Adds support for uploading a custom TensorFlow Object Detection model in the Manage page, which is especially useful for Blocks and OnBotJava users. Shows new Control Hub blink codes when the WiFi band is switched using the Control Hub's button (only possible on Control Hub OS 1.1.2) Adds new warnings which can be disabled in the Advanced RC Settings Mismatched app versions warning Unnecessary 2.4 GHz WiFi usage warning REV Hub is running outdated firmware (older than version 1.8.2) Adds support for Sony PS4 gamepad, and reworks how gamepads work on the Driver Station Removes preference which sets gamepad type based on driver position. Replaced with menu which allows specifying type for gamepads with unknown VID and PID Attempts to auto-detect gamepad type based on USB VID and PID If gamepad VID and PID is not known, use type specified by user for that VID and PID If gamepad VID and PID is not known AND the user has not specified a type for that VID and PID, an educated guess is made about how to map the gamepad Driver Station will now attempt to automatically recover from a gamepad disconnecting, and re-assign it to the position it was assigned to when it dropped If only one gamepad is assigned and it drops: it can be recovered If two gamepads are assigned, and have different VID/PID signatures, and only one drops: it will be recovered If two gamepads are assigned, and have different VID/PID signatures, and BOTH drop: both will be recovered If two gamepads are assigned, and have the same VID/PID signatures, and only one drops: it will be recovered If two gamepads are assigned, and have the same VID/PID signatures, and BOTH drop: neither will be recovered, because of the ambiguity of the gamepads when they re-appear on the USB bus. There is currently one known edge case: if there are two gamepads with the same VID/PID signature plugged in, but only one is assigned, and they BOTH drop, it's a 50-50 chance of which one will be chosen for automatic recovery to the assigned position: it is determined by whichever one is re-enumerated first by the USB bus controller. Adds landscape user interface to Driver Station New feature: practice timer with audio cues New feature (Control Hub only): wireless network connection strength indicator (0-5 bars) New feature (Control Hub only): tapping on the ping/channel display will switch to an alternate display showing radio RX dBm and link speed (tap again to switch back) The layout will NOT autorotate. You can switch the layout from the Driver Station's settings menu. Breaking changes Removes support for Android versions 4.4 through 5.1 (KitKat and Lollipop). The minSdkVersion is now 23. Removes the deprecated LinearOpMode methods waitOneFullHardwareCycle() and waitForNextHardwareCycle() Enhancements Handles RS485 address of Control Hub automatically The Control Hub is automatically given a reserved address Existing configuration files will continue to work All addresses in the range of 1-10 are still available for Expansion Hubs The Control Hub light will now normally be solid green, without blinking to indicate the address The Control Hub will not be shown on the Expansion Hub Address Change settings page Improves REV Hub firmware updater The user can now choose between all available firmware update files Version 1.8.2 of the REV Hub firmware is bundled into the Robot Controller app. Text was added to clarify that Expansion Hubs can only be updated via USB. Firmware update speed was reduced to improve reliability Allows REV Hub firmware to be updated directly from the Manage webpage Improves log viewer on Robot Controller Horizontal scrolling support (no longer word wrapped) Supports pinch-to-zoom Uses a monospaced font Error messages are highlighted New color scheme Attempts to force-stop a runaway/stuck OpMode without restarting the entire app Not all types of runaway conditions are stoppable, but if the user code attempts to talk to hardware during the runaway, the system should be able to capture it. Makes various tweaks to the Self Inspect screen Renames "OS version" entry to "Android version" Renames "WiFi Direct Name" to "WiFi Name" Adds Control Hub OS version, when viewing the report of a Control Hub Hides the airplane mode entry, when viewing the report of a Control Hub Removes check for ZTE Speed Channel Changer Shows firmware version for all Expansion and Control Hubs Reworks network settings portion of Manage page All network settings are now applied with a single click The WiFi Direct channel of phone-based Robot Controllers can now be changed from the Manage page WiFi channels are filtered by band (2.4 vs 5 GHz) and whether they overlap with other channels The current WiFi channel is pre-selected on phone-based Robot Controllers, and Control Hubs running OS 1.1.2 or later. On Control Hubs running OS 1.1.2 or later, you can choose to have the system automatically select a channel on the 5 GHz band Improves OnBotJava New light and dark themes replace the old themes (chaos, github, chrome,...) the new default theme is light and will be used when you first update to this version OnBotJava now has a tabbed editor Read-only offline mode Improves function of "exit" menu item on Robot Controller and Driver Station Now guaranteed to be fully stopped and unloaded from memory Shows a warning message if a LinearOpMode exists prematurely due to failure to monitor for the start condition Improves error message shown when the Driver Station and Robot Controller are incompatible with each other Driver Station OpMode Control Panel now disabled while a Restart Robot is in progress Disables advanced settings related to WiFi direct when the Robot Controller is a Control Hub. Tint phone battery icons on Driver Station when low/critical. Uses names "Control Hub Portal" and "Control Hub" (when appropriate) in new configuration files Improve I2C read performance Very large improvement on Control Hub; up to ~2x faster with small (e.g. 6 byte) reads Not as apparent on Expansion Hubs connected to a phone Update/refresh build infrastructure Update to 'androidx' support library from 'com.android.support:appcompat', which is end-of-life Update targetSdkVersion and compileSdkVersion to 28 Update Android Studio's Android plugin to latest Fix reported build timestamp in 'About' screen Add sample illustrating manual webcam use: ConceptWebcam Bug fixes Fixes SkyStone issue #248 Fixes SkyStone issue #232 and modifies bulk caching semantics to allow for cache-preserving MANUAL/AUTO transitions. Improves performance when REV 2M distance sensor is unplugged Improves readability of Toast messages on certain devices Allows a Driver Station to connect to a Robot Controller after another has disconnected Improves generation of fake serial numbers for UVC cameras which do not provide a real serial number Previously some devices would assign such cameras a serial of 0:0 and fail to open and start streaming Fixes ftc_app issue #638. Fixes a slew of bugs with the Vuforia camera monitor including: Fixes bug where preview could be displayed with a wonky aspect ratio Fixes bug where preview could be cut off in landscape Fixes bug where preview got totally messed up when rotating phone Fixes bug where crosshair could drift off target when using webcams Fixes issue in UVC driver on some devices (ftc_app 681) if streaming was started/stopped multiple times in a row Issue manifested as kernel panic on devices which do not have this kernel patch. On affected devices which do have the patch, the issue was manifest as simply a failure to start streaming. The Tech Team believes that the root cause of the issue is a bug in the Linux kernel XHCI driver. A workaround was implemented in the SDK UVC driver. Fixes bug in UVC driver where often half the frames from the camera would be dropped (e.g. only 15FPS delivered during a streaming session configured for 30FPS). Fixes issue where TensorFlow Object Detection would show results whose confidence was lower than the minimum confidence parameter. Fixes a potential exploitation issue of CVE-2019-11358 in OnBotJava Fixes changing the address of an Expansion Hub with additional Expansion Hubs connected to it Preserves the Control Hub's network connection when "Restart Robot" is selected Fixes issue where device scans would fail while the Robot was restarting Fix RenderScript usage Use androidx.renderscript variant: increased compatibility Use RenderScript in Java mode, not native: simplifies build Fixes webcam-frame-to-bitmap conversion problem: alpha channel wasn't being initialized, only R, G, & B Fixes possible arithmetic overflow in Deadline Fixes deadlock in Vuforia webcam support which could cause 5-second delays when stopping OpMode Version 5.4 (20200108-101156) Fixes SkyStone issue #88 Adds an inspection item that notes when a robot controller (Control Hub) is using the factory default password. Fixes SkyStone issue #61 Fixes SkyStone issue #142 Fixes ftc_app issue #417 by adding more current and voltage monitoring capabilities for REV Hubs. Fixes a crash sometimes caused by OnBotJava activity Improves OnBotJava autosave functionality ftc_app #738 Fixes system responsiveness issue when an Expansion Hub is disconnected Fixes issue where IMU initialization could prevent Op Modes from stopping Fixes issue where AndroidTextToSpeech.speak() would fail if it was called too early Adds telemetry.speak() methods and blocks, which cause the Driver Station (if also updated) to speak text Adds and improves Expansion Hub-related warnings Improves Expansion Hub low battery warning Displays the warning immediately after the hub reports it Specifies whether the condition is current or occurred temporarily during an OpMode run Displays which hubs reported low battery Displays warning when hub loses and regains power during an OpMode run Fixes the hub's LED pattern after this condition Displays warning when Expansion Hub is not responding to commands Specifies whether the condition is current or occurred temporarily during an OpMode run Clarifies warning when Expansion Hub is not present at startup Specifies that this condition requires a Robot Restart before the hub can be used. The hub light will now accurately reflect this state Improves logging and reduces log spam during these conditions Syncs the Control Hub time and timezone to a connected web browser programming the robot, if a Driver Station is not available. Adds bulk read functionality for REV Hubs A bulk caching mode must be set at the Hub level with LynxModule#setBulkCachingMode(). This applies to all relevant SDK hardware classes that reference that Hub. The following following Hub bulk caching modes are available: BulkCachingMode.OFF (default): All hardware calls operate as usual. Bulk data can read through LynxModule#getBulkData() and processed manually. BulkCachingMode.AUTO: Applicable hardware calls are served from a bulk read cache that is cleared/refreshed automatically to ensure identical commands don't hit the same cache. The cache can also be cleared manually with LynxModule#clearBulkCache(), although this is not recommended. (advanced users) BulkCachingMode.MANUAL: Same as BulkCachingMode.AUTO except the cache is never cleared automatically. To avoid getting stale data, the cache must be manually cleared at the beginning of each loop body or as the user deems appropriate. Removes PIDF Annotation values added in Rev 5.3 (to AndyMark, goBILDA and TETRIX motor configurations). The new motor types will still be available but their Default control behavior will revert back to Rev 5.2 Adds new ConceptMotorBulkRead sample Opmode to demonstrate and compare Motor Bulk-Read modes for reducing I/O latencies. Version 5.3 (20191004-112306) Fixes external USB/UVC webcam support Makes various bugfixes and improvements to Blocks page, including but not limited to: Many visual tweaks Browser zoom and window resize behave better Resizing the Java preview pane works better and more consistently across browsers The Java preview pane consistently gets scrollbars when needed The Java preview pane is hidden by default on phones Internet Explorer 11 should work Large dropdown lists display properly on lower res screens Disabled buttons are now visually identifiable as disabled A warning is shown if a user selects a TFOD sample, but their device is not compatible Warning messages in a Blocks op mode are now visible by default. Adds goBILDA 5201 and 5202 motors to Robot Configurator Adds PIDF Annotation values to AndyMark, goBILDA and TETRIX motor configurations. This has the effect of causing the RUN_USING_ENCODERS and RUN_TO_POSITION modes to use PIDF vs PID closed loop control on these motors. This should provide more responsive, yet stable, speed control. PIDF adds Feedforward control to the basic PID control loop. Feedforward is useful when controlling a motor's speed because it "anticipates" how much the control voltage must change to achieve a new speed set-point, rather than requiring the integrated error to change sufficiently. The PIDF values were chosen to provide responsive, yet stable, speed control on a lightly loaded motor. The more heavily a motor is loaded (drag or friction), the more noticable the PIDF improvement will be. Fixes startup crash on Android 10 Fixes ftc_app issue #712 (thanks to FROGbots-4634) Fixes ftc_app issue #542 Allows "A" and lowercase letters when naming device through RC and DS apps. Version 5.2 (20190905-083277) Fixes extra-wide margins on settings activities, and placement of the new configuration button Adds Skystone Vuforia image target data. Includes sample Skystone Vuforia Navigation op modes (Java). Includes sample Skystone Vuforia Navigation op modes (Blocks). Adds TensorFlow inference model (.tflite) for Skystone game elements. Includes sample Skystone TensorFlow op modes (Java). Includes sample Skystone TensorFlow op modes (Blocks). Removes older (season-specific) sample op modes. Includes 64-bit support (to comply with Google Play requirements). Protects against Stuck OpModes when a Restart Robot is requested. (Thanks to FROGbots-4634) (ftc_app issue #709) Blocks related changes: Fixes bug with blocks generated code when hardware device name is a java or javascript reserved word. Shows generated java code for blocks, even when hardware items are missing from the active configuration. Displays warning icon when outdated Vuforia and TensorFlow blocks are used (SkyStone issue #27) Version 5.1 (20190820-222104) Defines default PIDF parameters for the following motors: REV Core Hex Motor REV 20:1 HD Hex Motor REV 40:1 HD Hex Motor Adds back button when running on a device without a system back button (such as a Control Hub) Allows a REV Control Hub to update the firmware on a REV Expansion Hub via USB Fixes SkyStone issue #9 Fixes ftc_app issue #715 Prevents extra DS User clicks by filtering based on current state. Prevents incorrect DS UI state changes when receiving new OpMode list from RC Adds support for REV Color Sensor V3 Adds a manual-refresh DS Camera Stream for remotely viewing RC camera frames. To show the stream on the DS, initialize but do not run a stream-enabled opmode, select the Camera Stream option in the DS menu, and tap the image to refresh. This feature is automatically enabled when using Vuforia or TFOD—no additional RC configuration is required for typical use cases. To hide the stream, select the same menu item again. Note that gamepads are disabled and the selected opmode cannot be started while the stream is open as a safety precaution. To use custom streams, consult the API docs for CameraStreamServer#setSource and CameraStreamSource. Adds many Star Wars sounds to RobotController resources. Added SKYSTONE Sounds Chooser Sample Program. Switches out startup, connect chimes, and error/warning sounds for Star Wars sounds Updates OnBot Java to use a WebSocket for communication with the robot The OnBot Java page no longer has to do a full refresh when a user switches from editing one file to another Known issues: Camera Stream The Vuforia camera stream inherits the issues present in the phone preview (namely ftc_app issue #574). This problem does not affect the TFOD camera stream even though it receives frames from Vuforia. The orientation of the stream frames may not always match the phone preview. For now, these frames may be rotated manually via a custom CameraStreamSource if desired. OnBotJava Browser back button may not always work correctly It's possible for a build to be queued, but not started. The OnBot Java build console will display a warning if this occurs. A user might not realize they are editing a different file if the user inadvertently switches from one file to another since this switch is now seamless. The name of the currently open file is displayed in the browser tab. Version 5.0 (built on 19.06.14) Support for the REV Robotics Control Hub. Adds a Java preview pane to the Blocks editor. Adds a new offline export feature to the Blocks editor. Display wifi channel in Network circle on Driver Station. Adds calibration for Logitech C270 Updates build tooling and target SDK. Compliance with Google's permissions infrastructure (Required after build tooling update). Keep Alives to mitigate the Motorola wifi scanning problem. Telemetry substitute no longer necessary. Improves Vuforia error reporting. Fixes ftctechnh/ftc_app issues 621, 713. Miscellaneous bug fixes and improvements. Version 4.3 (built on 18.10.31) Includes missing TensorFlow-related libraries and files. Version 4.2 (built on 18.10.30) Includes fix to avoid deadlock situation with WatchdogMonitor which could result in USB communication errors. Comm error appeared to require that user disconnect USB cable and restart the Robot Controller app to recover. robotControllerLog.txt would have error messages that included the words "E RobotCore: lynx xmit lock: #### abandoning lock:" Includes fix to correctly list the parent module address for a REV Robotics Expansion Hub in a configuration (.xml) file. Bug in versions 4.0 and 4.1 would incorrect list the address module for a parent REV Robotics device as "1". If the parent module had a higher address value than the daisy-chained module, then this bug would prevent the Robot Controller from communicating with the downstream Expansion Hub. Added requirement for ACCESS_COARSE_LOCATION to allow a Driver Station running Android Oreo to scan for Wi-Fi Direct devices. Added google() repo to build.gradle because aapt2 must be downloaded from the google() repository beginning with version 3.2 of the Android Gradle Plugin. Important Note: Android Studio users will need to be connected to the Internet the first time build the ftc_app project. Internet connectivity is required for the first build so the appropriate files can be downloaded from the Google repository. Users should not need to be connected to the Internet for subsequent builds. This should also fix buid issue where Android Studio would complain that it "Could not find com.android.tools.lint:lint-gradle:26.1.4" (or similar). Added support for REV Spark Mini motor controller as part of the configuration menu for a servo/PWM port on the REV Expansion Hub. Provide examples for playing audio files in an Op Mode. Block Development Tool Changes Includes a fix for a problem with the Velocity blocks that were reported in the FTC Technology forum (Blocks Programming subforum). Change the "Save completed successfully." message to a white color so it will contrast with a green background. Fixed the "Download image" feature so it will work if there are text blocks in the op mode. Introduce support for Google's TensorFlow Lite technology for object detetion for 2018-2019 game. TensorFlow lite can recognize Gold Mineral and Silver Mineral from 2018-2019 game. Example Java and Block op modes are included to show how to determine the relative position of the gold block (left, center, right). Version 4.1 (released on 18.09.24) Changes include: Fix to prevent crash when deprecated configuration annotations are used. Change to allow FTC Robot Controller APK to be auto-updated using FIRST Global Control Hub update scripts. Removed samples for non supported / non legal hardware. Improvements to Telemetry.addData block with "text" socket. Updated Blocks sample op mode list to include Rover Ruckus Vuforia example. Update SDK library version number. Version 4.0 (released on 18.09.12) Changes include: Initial support for UVC compatible cameras If UVC camera has a unique serial number, RC will detect and enumerate by serial number. If UVC camera lacks a unique serial number, RC will only support one camera of that type connected. Calibration settings for a few cameras are included (see TeamCode/src/main/res/xml/teamwebcamcalibrations.xml for details). User can upload calibration files from Program and Manage web interface. UVC cameras seem to draw a fair amount of electrical current from the USB bus. This does not appear to present any problems for the REV Robotics Control Hub. This does seem to create stability problems when using some cameras with an Android phone-based Robot Controller. FTC Tech Team is investigating options to mitigate this issue with the phone-based Robot Controllers. Updated sample Vuforia Navigation and VuMark Op Modes to demonstrate how to use an internal phone-based camera and an external UVC webcam. Support for improved motor control. REV Robotics Expansion Hub firmware 1.8 and greater will support a feed forward mechanism for closed loop motor control. FTC SDK has been modified to support PIDF coefficients (proportional, integral, derivative, and feed forward). FTC Blocks development tool modified to include PIDF programming blocks. Deprecated older PID-related methods and variables. REV's 1.8.x PIDF-related changes provide a more linear and accurate way to control a motor. Wireless Added 5GHz support for wireless channel changing for those devices that support it. Tested with Moto G5 and E4 phones. Also tested with other (currently non-approved) phones such as Samsung Galaxy S8. Improved Expansion Hub firmware update support in Robot Controller app Changes to make the system more robust during the firmware update process (when performed through Robot Controller app). User no longer has to disconnect a downstream daisy-chained Expansion Hub when updating an Expansion Hub's firmware. If user is updating an Expansion Hub's firmware through a USB connection, he/she does not have to disconnect RS485 connection to other Expansion Hubs. The user still must use a USB connection to update an Expansion Hub's firmware. The user cannot update the Expansion Hub firmware for a downstream device that is daisy chained through an RS485 connection. If an Expansion Hub accidentally gets "bricked" the Robot Controller app is now more likely to recognize the Hub when it scans the USB bus. Robot Controller app should be able to detect an Expansion Hub, even if it accidentally was bricked in a previous update attempt. Robot Controller app should be able to install the firmware onto the Hub, even if if accidentally was bricked in a previous update attempt. Resiliency FTC software can detect and enable an FTDI reset feature that is available with REV Robotics v1.8 Expansion Hub firmware and greater. When enabled, the Expansion Hub can detect if it hasn't communicated with the Robot Controller over the FTDI (USB) connection. If the Hub hasn't heard from the Robot Controller in a while, it will reset the FTDI connection. This action helps system recover from some ESD-induced disruptions. Various fixes to improve reliability of FTC software. Blocks Fixed errors with string and list indices in blocks export to java. Support for USB connected UVC webcams. Refactored optimized Blocks Vuforia code to support Rover Ruckus image targets. Added programming blocks to support PIDF (proportional, integral, derivative and feed forward) motor control. Added formatting options (under Telemetry and Miscellaneous categories) so user can set how many decimal places to display a numerical value. Support to play audio files (which are uploaded through Blocks web interface) on Driver Station in addition to the Robot Controller. Fixed bug with Download Image of Blocks feature. Support for REV Robotics Blinkin LED Controller. Support for REV Robotics 2m Distance Sensor. Added support for a REV Touch Sensor (no longer have to configure as a generic digital device). Added blocks for DcMotorEx methods. These are enhanced methods that you can use when supported by the motor controller hardware. The REV Robotics Expansion Hub supports these enhanced methods. Enhanced methods include methods to get/set motor velocity (in encoder pulses per second), get/set PIDF coefficients, etc.. Modest Improvements in Logging Decrease frequency of battery checker voltage statements. Removed non-FTC related log statements (wherever possible). Introduced a "Match Logging" feature. Under "Settings" a user can enable/disable this feature (it's disabled by default). If enabled, user provides a "Match Number" through the Driver Station user interface (top of the screen). The Match Number is used to create a log file specifically with log statements from that particular Op Mode run. Match log files are stored in /sdcard/FIRST/matlogs on the Robot Controller. Once an op mode run is complete, the Match Number is cleared. This is a convenient way to create a separate match log with statements only related to a specific op mode run. New Devices Support for REV Robotics Blinkin LED Controller. Support for REV Robotics 2m Distance Sensor. Added configuration option for REV 20:1 HD Hex Motor. Added support for a REV Touch Sensor (no longer have to configure as a generic digital device). Miscellaneous Fixed some errors in the definitions for acceleration and velocity in our javadoc documentation. Added ability to play audio files on Driver Station When user is configuring an Expansion Hub, the LED on the Expansion Hub will change blink pattern (purple-cyan) to indicate which Hub is currently being configured. Renamed I2cSensorType to I2cDeviceType. Added an external sample Op Mode that demonstrates localization using 2018-2019 (Rover Ruckus presented by QualComm) Vuforia targets. Added an external sample Op Mode that demonstrates how to use the REV Robotics 2m Laser Distance Sensor. Added an external sample Op Mode that demonstrates how to use the REV Robotics Blinkin LED Controller. Re-categorized external Java sample Op Modes to "TeleOp" instead of "Autonomous". Known issues: Initial support for UVC compatible cameras UVC cameras seem to draw significant amount of current from the USB bus. This does not appear to present any problems for the REV Robotics Control Hub. This does seem to create stability problems when using some cameras with an Android phone-based Robot Controller. FTC Tech Team is investigating options to mitigate this issue with the phone-based Robot Controllers. There might be a possible deadlock which causes the RC to become unresponsive when using a UVC webcam with a Nougat Android Robot Controller. Wireless When user selects a wireless channel, this channel does not necessarily persist if the phone is power cycled. Tech Team is hoping to eventually address this issue in a future release. Issue has been present since apps were introduced (i.e., it is not new with the v4.0 release). Wireless channel is not currently displayed for WiFi Direct connections. Miscellaneous The blink indication feature that shows which Expansion Hub is currently being configured does not work for a newly created configuration file. User has to first save a newly created configuration file and then close and re-edit the file in order for blink indicator to work. Version 3.6 (built on 17.12.18) Changes include: Blocks Changes Uses updated Google Blockly software to allow users to edit their op modes on Apple iOS devices (including iPad and iPhone). Improvement in Blocks tool to handle corrupt op mode files. Autonomous op modes should no longer get switched back to tele-op after re-opening them to be edited. The system can now detect type mismatches during runtime and alert the user with a message on the Driver Station. Updated javadoc documentation for setPower() method to reflect correct range of values (-1 to +1). Modified VuforiaLocalizerImpl to allow for user rendering of frames Added a user-overrideable onRenderFrame() method which gets called by the class's renderFrame() method. Version 3.5 (built on 17.10.30) Changes with version 3.5 include: Introduced a fix to prevent random op mode stops, which can occur after the Robot Controller app has been paused and then resumed (for example, when a user temporarily turns off the display of the Robot Controller phone, and then turns the screen back on). Introduced a fix to prevent random op mode stops, which were previously caused by random peer disconnect events on the Driver Station. Fixes issue where log files would be closed on pause of the RC or DS, but not re-opened upon resume. Fixes issue with battery handler (voltage) start/stop race. Fixes issue where Android Studio generated op modes would disappear from available list in certain situations. Fixes problem where OnBot Java would not build on REV Robotics Control Hub. Fixes problem where OnBot Java would not build if the date and time on the Robot Controller device was "rewound" (set to an earlier date/time). Improved error message on OnBot Java that occurs when renaming a file fails. Removed unneeded resources from android.jar binaries used by OnBot Java to reduce final size of Robot Controller app. Added MR_ANALOG_TOUCH_SENSOR block to Blocks Programming Tool. Version 3.4 (built on 17.09.06) Changes with version 3.4 include: Added telemetry.update() statement for BlankLinearOpMode template. Renamed sample Block op modes to be more consistent with Java samples. Added some additional sample Block op modes. Reworded OnBot Java readme slightly. Version 3.3 (built on 17.09.04) This version of the software includes improves for the FTC Blocks Programming Tool and the OnBot Java Programming Tool. Changes with verion 3.3 include: Android Studio ftc_app project has been updated to use Gradle Plugin 2.3.3. Android Studio ftc_app project is already using gradle 3.5 distribution. Robot Controller log has been renamed to /sdcard/RobotControllerLog.txt (note that this change was actually introduced w/ v3.2). Improvements in I2C reliability. Optimized I2C read for REV Expansion Hub, with v1.7 firmware or greater. Updated all external/samples (available through OnBot and in Android project folder). Vuforia Added support for VuMarks that will be used for the 2017-2018 season game. Blocks Update to latest Google Blockly release. Sample op modes can be selected as a template when creating new op mode. Fixed bug where the blocks would disappear temporarily when mouse button is held down. Added blocks for Range.clip and Range.scale. User can now disable/enable Block op modes. Fix to prevent occasional Blocks deadlock. OnBot Java Significant improvements with autocomplete function for OnBot Java editor. Sample op modes can be selected as a template when creating new op mode. Fixes and changes to complete hardware setup feature. Updated (and more useful) onBot welcome message. Known issues: Android Studio After updating to the new v3.3 Android Studio project folder, if you get error messages indicating "InvalidVirtualFileAccessException" then you might need to do a File->Invalidate Caches / Restart to clear the error. OnBot Java Sometimes when you push the build button to build all op modes, the RC returns an error message that the build failed. If you press the build button a second time, the build typically suceeds. Version 3.2 (built on 17.08.02) This version of the software introduces the "OnBot Java" Development Tool. Similar to the FTC Blocks Development Tool, the FTC OnBot Java Development Tool allows a user to create, edit and build op modes dynamically using only a Javascript-enabled web browser. The OnBot Java Development Tool is an integrated development environment (IDE) that is served up by the Robot Controller. Op modes are created and edited using a Javascript-enabled browser (Google Chromse is recommended). Op modes are saved on the Robot Controller Android device directly. The OnBot Java Development Tool provides a Java programming environment that does NOT need Android Studio. Changes with version 3.2 include: Enhanced web-based development tools Introduction of OnBot Java Development Tool. Web-based programming and management features are "always on" (user no longer needs to put Robot Controller into programming mode). Web-based management interface (where user can change Robot Controller name and also easily download Robot Controller log file). OnBot Java, Blocks and Management features available from web based interface. Blocks Programming Development Tool: Changed "LynxI2cColorRangeSensor" block to "REV Color/range sensor" block. Fixed tooltip for ColorSensor.isLightOn block. Added blocks for ColorSensor.getNormalizedColors and LynxI2cColorRangeSensor.getNormalizedColors. Added example op modes for digital touch sensor and REV Robotics Color Distance sensor. User selectable color themes. Includes many minor enhancements and fixes (too numerous to list). Known issues: Auto complete function is incomplete and does not support the following (for now): Access via this keyword Access via super keyword Members of the super cloass, not overridden by the class Any methods provided in the current class Inner classes Can't handle casted objects Any objects coming from an parenthetically enclosed expression Version 3.10 (built on 17.05.09) This version of the software provides support for the REV Robotics Expansion Hub. This version also includes improvements in the USB communication layer in an effort to enhance system resiliency. If you were using a 2.x version of the software previously, updating to version 3.1 requires that you also update your Driver Station software in addition to updating the Robot Controller software. Also note that in version 3.10 software, the setMaxSpeed and getMaxSpeed methods are no longer available (not deprecated, they have been removed from the SDK). Also note that the the new 3.x software incorporates motor profiles that a user can select as he/she configures the robot. Changes include: Blocks changes Added VuforiaTrackableDefaultListener.getPose and Vuforia.trackPose blocks. Added optimized blocks support for Vuforia extended tracking. Added atan2 block to the math category. Added useCompetitionFieldTargetLocations parameter to Vuforia.initialize block. If set to false, the target locations are placed at (0,0,0) with target orientation as specified in https://github.com/gearsincorg/FTCVuforiaDemo/blob/master/Robot_Navigation.java tutorial op mode. Incorporates additional improvements to USB comm layer to improve system resiliency (to recover from a greater number of communication disruptions). Additional Notes Regarding Version 3.00 (built on 17.04.13) In addition to the release changes listed below (see section labeled "Version 3.00 (built on 17.04.013)"), version 3.00 has the following important changes: Version 3.00 software uses a new version of the FTC Robocol (robot protocol). If you upgrade to v3.0 on the Robot Controller and/or Android Studio side, you must also upgrade the Driver Station software to match the new Robocol. Version 3.00 software removes the setMaxSpeed and getMaxSpeed methods from the DcMotor class. If you have an op mode that formerly used these methods, you will need to remove the references/calls to these methods. Instead, v3.0 provides the max speed information through the use of motor profiles that are selected by the user during robot configuration. Version 3.00 software currently does not have a mechanism to disable extra i2c sensors. We hope to re-introduce this function with a release in the near future. Version 3.00 (built on 17.04.13) *** Use this version of the software at YOUR OWN RISK!!! *** This software is being released as an "alpha" version. Use this version at your own risk! This pre-release software contains SIGNIFICANT changes, including changes to the Wi-Fi Direct pairing mechanism, rewrites of the I2C sensor classes, changes to the USB/FTDI layer, and the introduction of support for the REV Robotics Expansion Hub and the REV Robotics color-range-light sensor. These changes were implemented to improve the reliability and resiliency of the FTC control system. Please note, however, that version 3.00 is considered "alpha" code. This code is being released so that the FIRST community will have an opportunity to test the new REV Expansion Hub electronics module when it becomes available in May. The developers do not recommend using this code for critical applications (i.e., competition use). *** Use this version of the software at YOUR OWN RISK!!! *** Changes include: Major rework of sensor-related infrastructure. Includes rewriting sensor classes to implement synchronous I2C communication. Fix to reset Autonomous timer back to 30 seconds. Implementation of specific motor profiles for approved 12V motors (includes Tetrix, AndyMark, Matrix and REV models). Modest improvements to enhance Wi-Fi P2P pairing. Fixes telemetry log addition race. Publishes all the sources (not just a select few). Includes Block programming improvements Addition of optimized Vuforia blocks. Auto scrollbar to projects and sounds pages. Fixed blocks paste bug. Blocks execute after while-opModeIsActive loop (to allow for cleanup before exiting op mode). Added gyro integratedZValue block. Fixes bug with projects page for Firefox browser. Added IsSpeaking block to AndroidTextToSpeech. Implements support for the REV Robotics Expansion Hub Implements support for integral REV IMU (physically installed on I2C bus 0, uses same Bosch BNO055 9 axis absolute orientation sensor as Adafruit 9DOF abs orientation sensor). - Implements support for REV color/range/light sensor. Provides support to update Expansion Hub firmware through FTC SDK. Detects REV firmware version and records in log file. Includes support for REV Control Hub (note that the REV Control Hub is not yet approved for FTC use). Implements FTC Blocks programming support for REV Expansion Hub and sensor hardware. Detects and alerts when I2C device disconnect. Version 2.62 (built on 17.01.07) Added null pointer check before calling modeToByte() in finishModeSwitchIfNecessary method for ModernRoboticsUsbDcMotorController class. Changes to enhance Modern Robotics USB protocol robustness. Version 2.61 (released on 16.12.19) Blocks Programming mode changes: Fix to correct issue when an exception was thrown because an OpticalDistanceSensor object appears twice in the hardware map (the second time as a LightSensor). Version 2.6 (released on 16.12.16) Fixes for Gyro class: Improve (decrease) sensor refresh latency. fix isCalibrating issues. Blocks Programming mode changes: Blocks now ignores a device in the configuration xml if the name is empty. Other devices work in configuration work fine. Version 2.5 (internal release on released on 16.12.13) Blocks Programming mode changes: Added blocks support for AdafruitBNO055IMU. Added Download Op Mode button to FtcBocks.html. Added support for copying blocks in one OpMode and pasting them in an other OpMode. The clipboard content is stored on the phone, so the programming mode server must be running. Modified Utilities section of the toolbox. In Programming Mode, display information about the active connections. Fixed paste location when workspace has been scrolled. Added blocks support for the android Accelerometer. Fixed issue where Blocks Upload Op Mode truncated name at first dot. Added blocks support for Android SoundPool. Added type safety to blocks for Acceleration. Added type safety to blocks for AdafruitBNO055IMU.Parameters. Added type safety to blocks for AnalogInput. Added type safety to blocks for AngularVelocity. Added type safety to blocks for Color. Added type safety to blocks for ColorSensor. Added type safety to blocks for CompassSensor. Added type safety to blocks for CRServo. Added type safety to blocks for DigitalChannel. Added type safety to blocks for ElapsedTime. Added type safety to blocks for Gamepad. Added type safety to blocks for GyroSensor. Added type safety to blocks for IrSeekerSensor. Added type safety to blocks for LED. Added type safety to blocks for LightSensor. Added type safety to blocks for LinearOpMode. Added type safety to blocks for MagneticFlux. Added type safety to blocks for MatrixF. Added type safety to blocks for MrI2cCompassSensor. Added type safety to blocks for MrI2cRangeSensor. Added type safety to blocks for OpticalDistanceSensor. Added type safety to blocks for Orientation. Added type safety to blocks for Position. Added type safety to blocks for Quaternion. Added type safety to blocks for Servo. Added type safety to blocks for ServoController. Added type safety to blocks for Telemetry. Added type safety to blocks for Temperature. Added type safety to blocks for TouchSensor. Added type safety to blocks for UltrasonicSensor. Added type safety to blocks for VectorF. Added type safety to blocks for Velocity. Added type safety to blocks for VoltageSensor. Added type safety to blocks for VuforiaLocalizer.Parameters. Added type safety to blocks for VuforiaTrackable. Added type safety to blocks for VuforiaTrackables. Added type safety to blocks for enums in AdafruitBNO055IMU.Parameters. Added type safety to blocks for AndroidAccelerometer, AndroidGyroscope, AndroidOrientation, and AndroidTextToSpeech. Version 2.4 (released on 16.11.13) Fix to avoid crashing for nonexistent resources. Blocks Programming mode changes: Added blocks to support OpenGLMatrix, MatrixF, and VectorF. Added blocks to support AngleUnit, AxesOrder, AxesReference, CameraDirection, CameraMonitorFeedback, DistanceUnit, and TempUnit. Added blocks to support Acceleration. Added blocks to support LinearOpMode.getRuntime. Added blocks to support MagneticFlux and Position. Fixed typos. Made blocks for ElapsedTime more consistent with other objects. Added blocks to support Quaternion, Velocity, Orientation, AngularVelocity. Added blocks to support VuforiaTrackables, VuforiaTrackable, VuforiaLocalizer, VuforiaTrackableDefaultListener. Fixed a few blocks. Added type checking to new blocks. Updated to latest blockly. Added default variable blocks to navigation and matrix blocks. Fixed toolbox entry for openGLMatrix_rotation_withAxesArgs. When user downloads Blocks-generated op mode, only the .blk file is downloaded. When user uploads Blocks-generated op mode (.blk file), Javascript code is auto generated. Added DbgLog support. Added logging when a blocks file is read/written. Fixed bug to properly render blocks even if missing devices from configuration file. Added support for additional characters (not just alphanumeric) for the block file names (for download and upload). Added support for OpMode flavor (“Autonomous” or “TeleOp”) and group. Changes to Samples to prevent tutorial issues. Incorporated suggested changes from public pull 216 (“Replace .. paths”). Remove Servo Glitches when robot stopped. if user hits “Cancels” when editing a configuration file, clears the unsaved changes and reverts to original unmodified configuration. Added log info to help diagnose why the Robot Controller app was terminated (for example, by watch dog function). Added ability to transfer log from the controller. Fixed inconsistency for AngularVelocity Limit unbounded growth of data for telemetry. If user does not call telemetry.update() for LinearOpMode in a timely manner, data added for telemetry might get lost if size limit is exceeded. Version 2.35 (released on 16.10.06) Blockly programming mode - Removed unnecesary idle() call from blocks for new project. Version 2.30 (released on 16.10.05) Blockly programming mode: Mechanism added to save Blockly op modes from Programming Mode Server onto local device To avoid clutter, blocks are displayed in categorized folders Added support for DigitalChannel Added support for ModernRoboticsI2cCompassSensor Added support for ModernRoboticsI2cRangeSensor Added support for VoltageSensor Added support for AnalogInput Added support for AnalogOutput Fix for CompassSensor setMode block Vuforia Fix deadlock / make camera data available while Vuforia is running. Update to Vuforia 6.0.117 (recommended by Vuforia and Google to close security loophole). Fix for autonomous 30 second timer bug (where timer was in effect, even though it appeared to have timed out). opModeIsActive changes to allow cleanup after op mode is stopped (with enforced 2 second safety timeout). Fix to avoid reading i2c twice. Updated sample Op Modes. Improved logging and fixed intermittent freezing. Added digital I/O sample. Cleaned up device names in sample op modes to be consistent with Pushbot guide. Fix to allow use of IrSeekerSensorV3. Version 2.20 (released on 16.09.08) Support for Modern Robotics Compass Sensor. Support for Modern Robotics Range Sensor. Revise device names for Pushbot templates to match the names used in Pushbot guide. Fixed bug so that IrSeekerSensorV3 device is accessible as IrSeekerSensor in hardwareMap. Modified computer vision code to require an individual Vuforia license (per legal requirement from PTC). Minor fixes. Blockly enhancements: Support for Voltage Sensor. Support for Analog Input. Support for Analog Output. Support for Light Sensor. Support for Servo Controller. Version 2.10 (released on 16.09.03) Support for Adafruit IMU. Improvements to ModernRoboticsI2cGyro class Block on reset of z axis. isCalibrating() returns true while gyro is calibration. Updated sample gyro program. Blockly enhancements support for android.graphics.Color. added support for ElapsedTime. improved look and legibility of blocks. support for compass sensor. support for ultrasonic sensor. support for IrSeeker. support for LED. support for color sensor. support for CRServo prompt user to configure robot before using programming mode. Provides ability to disable audio cues. various bug fixes and improvements. Version 2.00 (released on 16.08.19) This is the new release for the upcoming 2016-2017 FIRST Tech Challenge Season. Channel change is enabled in the FTC Robot Controller app for Moto G 2nd and 3rd Gen phones. Users can now use annotations to register/disable their Op Modes. Changes in the Android SDK, JDK and build tool requirements (minsdk=19, java 1.7, build tools 23.0.3). Standardized units in analog input. Cleaned up code for existing analog sensor classes. setChannelMode and getChannelMode were REMOVED from the DcMotorController class. This is important - we no longer set the motor modes through the motor controller. setMode and getMode were added to the DcMotor class. ContinuousRotationServo class has been added to the FTC SDK. Range.clip() method has been overloaded so it can support this operation for int, short and byte integers. Some changes have been made (new methods added) on how a user can access items from the hardware map. Users can now set the zero power behavior for a DC motor so that the motor will brake or float when power is zero. Prototype Blockly Programming Mode has been added to FTC Robot Controller. Users can place the Robot Controller into this mode, and then use a device (such as a laptop) that has a Javascript enabled browser to write Blockly-based Op Modes directly onto the Robot Controller. Users can now configure the robot remotely through the FTC Driver Station app. Android Studio project supports Android Studio 2.1.x and compile SDK Version 23 (Marshmallow). Vuforia Computer Vision SDK integrated into FTC SDK. Users can use sample vision targets to get localization information on a standard FTC field. Project structure has been reorganized so that there is now a TeamCode package that users can use to place their local/custom Op Modes into this package. Inspection function has been integrated into the FTC Robot Controller and Driver Station Apps (Thanks Team HazMat… 9277 & 10650!). Audio cues have been incorporated into FTC SDK. Swap mechanism added to FTC Robot Controller configuration activity. For example, if you have two motor controllers on a robot, and you misidentified them in your configuration file, you can use the Swap button to swap the devices within the configuration file (so you do not have to manually re-enter in the configuration info for the two devices). Fix mechanism added to all user to replace an electronic module easily. For example, suppose a servo controller dies on your robot. You replace the broken module with a new module, which has a different serial number from the original servo controller. You can use the Fix button to automatically reconfigure your configuration file to use the serial number of the new module. Improvements made to fix resiliency and responsiveness of the system. For LinearOpMode the user now must for a telemetry.update() to update the telemetry data on the driver station. This update() mechanism ensures that the driver station gets the updated data properly and at the same time. The Auto Configure function of the Robot Controller is now template based. If there is a commonly used robot configuration, a template can be created so that the Auto Configure mechanism can be used to quickly configure a robot of this type. The logic to detect a runaway op mode (both in the LinearOpMode and OpMode types) and to abort the run, then auto recover has been improved/implemented. Fix has been incorporated so that Logitech F310 gamepad mappings will be correct for Marshmallow users. Release 16.07.08 For the ftc_app project, the gradle files have been modified to support Android Studio 2.1.x. Release 16.03.30 For the MIT App Inventor, the design blocks have new icons that better represent the function of each design component. Some changes were made to the shutdown logic to ensure the robust shutdown of some of our USB services. A change was made to LinearOpMode so as to allow a given instance to be executed more than once, which is required for the App Inventor. Javadoc improved/updated. Release 16.03.09 Changes made to make the FTC SDK synchronous (significant change!) waitOneFullHardwareCycle() and waitForNextHardwareCycle() are no longer needed and have been deprecated. runOpMode() (for a LinearOpMode) is now decoupled from the system's hardware read/write thread. loop() (for an OpMode) is now decoupled from the system's hardware read/write thread. Methods are synchronous. For example, if you call setMode(DcMotorController.RunMode.RESET_ENCODERS) for a motor, the encoder is guaranteed to be reset when the method call is complete. For legacy module (NXT compatible), user no longer has to toggle between read and write modes when reading from or writing to a legacy device. Changes made to enhance reliability/robustness during ESD event. Changes made to make code thread safe. Debug keystore added so that user-generated robot controller APKs will all use the same signed key (to avoid conflicts if a team has multiple developer laptops for example). Firmware version information for Modern Robotics modules are now logged. Changes made to improve USB comm reliability and robustness. Added support for voltage indicator for legacy (NXT-compatible) motor controllers. Changes made to provide auto stop capabilities for op modes. A LinearOpMode class will stop when the statements in runOpMode() are complete. User does not have to push the stop button on the driver station. If an op mode is stopped by the driver station, but there is a run away/uninterruptible thread persisting, the app will log an error message then force itself to crash to stop the runaway thread. Driver Station UI modified to display lowest measured voltage below current voltage (12V battery). Driver Station UI modified to have color background for current voltage (green=good, yellow=caution, red=danger, extremely low voltage). javadoc improved (edits and additional classes). Added app build time to About activity for driver station and robot controller apps. Display local IP addresses on Driver Station About activity. Added I2cDeviceSynchImpl. Added I2cDeviceSync interface. Added seconds() and milliseconds() to ElapsedTime for clarity. Added getCallbackCount() to I2cDevice. Added missing clearI2cPortActionFlag. Added code to create log messages while waiting for LinearOpMode shutdown. Fix so Wifi Direct Config activity will no longer launch multiple times. Added the ability to specify an alternate i2c address in software for the Modern Robotics gyro. Release 16.02.09 Improved battery checker feature so that voltage values get refreshed regularly (every 250 msec) on Driver Station (DS) user interface. Improved software so that Robot Controller (RC) is much more resilient and “self-healing” to USB disconnects: If user attempts to start/restart RC with one or more module missing, it will display a warning but still start up. When running an op mode, if one or more modules gets disconnected, the RC & DS will display warnings,and robot will keep on working in spite of the missing module(s). If a disconnected module gets physically reconnected the RC will auto detect the module and the user will regain control of the recently connected module. Warning messages are more helpful (identifies the type of module that’s missing plus its USB serial number). Code changes to fix the null gamepad reference when users try to reference the gamepads in the init() portion of their op mode. NXT light sensor output is now properly scaled. Note that teams might have to readjust their light threshold values in their op modes. On DS user interface, gamepad icon for a driver will disappear if the matching gamepad is disconnected or if that gamepad gets designated as a different driver. Robot Protocol (ROBOCOL) version number info is displayed in About screen on RC and DS apps. Incorporated a display filter on pairing screen to filter out devices that don’t use the “-“ format. This filter can be turned off to show all WiFi Direct devices. Updated text in License file. Fixed formatting error in OpticalDistanceSensor.toString(). Fixed issue on with a blank (“”) device name that would disrupt WiFi Direct Pairing. Made a change so that the WiFi info and battery info can be displayed more quickly on the DS upon connecting to RC. Improved javadoc generation. Modified code to make it easier to support language localization in the future. Release 16.01.04 Updated compileSdkVersion for apps Prevent Wifi from entering power saving mode removed unused import from driver station Corrrected "Dead zone" joystick code. LED.getDeviceName and .getConnectionInfo() return null apps check for ROBOCOL_VERSION mismatch Fix for Telemetry also has off-by-one errors in its data string sizing / short size limitations error User telemetry output is sorted. added formatting variants to DbgLog and RobotLog APIs code modified to allow for a long list of op mode names. changes to improve thread safety of RobocolDatagramSocket Fix for "missing hardware leaves robot controller disconnected from driver station" error fix for "fast tapping of Init/Start causes problems" (toast is now only instantiated on UI thread). added some log statements for thread life cycle. moved gamepad reset logic inside of initActiveOpMode() for robustness changes made to mitigate risk of race conditions on public methods. changes to try and flag when WiFi Direct name contains non-printable characters. fix to correct race condition between .run() and .close() in ReadWriteRunnableStandard. updated FTDI driver made ReadWriteRunnableStanard interface public. fixed off-by-one errors in Command constructor moved specific hardware implmentations into their own package. moved specific gamepad implemnatations to the hardware library. changed LICENSE file to new BSD version. fixed race condition when shutting down Modern Robotics USB devices. methods in the ColorSensor classes have been synchronized. corrected isBusy() status to reflect end of motion. corrected "back" button keycode. the notSupported() method of the GyroSensor class was changed to protected (it should not be public). Release 15.11.04.001 Added Support for Modern Robotics Gyro. The GyroSensor class now supports the MR Gyro Sensor. Users can access heading data (about Z axis) Users can also access raw gyro data (X, Y, & Z axes). Example MRGyroTest.java op mode included. Improved error messages More descriptive error messages for exceptions in user code. Updated DcMotor API Enable read mode on new address in setI2cAddress Fix so that driver station app resets the gamepads when switching op modes. USB-related code changes to make USB comm more responsive and to display more explicit error messages. Fix so that USB will recover properly if the USB bus returns garbage data. Fix USB initializtion race condition. Better error reporting during FTDI open. More explicit messages during USB failures. Fixed bug so that USB device is closed if event loop teardown method was not called. Fixed timer UI issue Fixed duplicate name UI bug (Legacy Module configuration). Fixed race condition in EventLoopManager. Fix to keep references stable when updating gamepad. For legacy Matrix motor/servo controllers removed necessity of appending "Motor" and "Servo" to controller names. Updated HT color sensor driver to use constants from ModernRoboticsUsbLegacyModule class. Updated MR color sensor driver to use constants from ModernRoboticsUsbDeviceInterfaceModule class. Correctly handle I2C Address change in all color sensors Updated/cleaned up op modes. Updated comments in LinearI2cAddressChange.java example op mode. Replaced the calls to "setChannelMode" with "setMode" (to match the new of the DcMotor method). Removed K9AutoTime.java op mode. Added MRGyroTest.java op mode (demonstrates how to use MR Gyro Sensor). Added MRRGBExample.java op mode (demonstrates how to use MR Color Sensor). Added HTRGBExample.java op mode (demonstrates how to use HT legacy color sensor). Added MatrixControllerDemo.java (demonstrates how to use legacy Matrix controller). Updated javadoc documentation. Updated release .apk files for Robot Controller and Driver Station apps. Release 15.10.06.002 Added support for Legacy Matrix 9.6V motor/servo controller. Cleaned up build.gradle file. Minor UI and bug fixes for driver station and robot controller apps. Throws error if Ultrasonic sensor (NXT) is not configured for legacy module port 4 or 5. Release 15.08.03.001 New user interfaces for FTC Driver Station and FTC Robot Controller apps. An init() method is added to the OpMode class. For this release, init() is triggered right before the start() method. Eventually, the init() method will be triggered when the user presses an "INIT" button on driver station. The init() and loop() methods are now required (i.e., need to be overridden in the user's op mode). The start() and stop() methods are optional. A new LinearOpMode class is introduced. Teams can use the LinearOpMode mode to create a linear (not event driven) program model. Teams can use blocking statements like Thread.sleep() within a linear op mode. The API for the Legacy Module and Core Device Interface Module have been updated. Support for encoders with the Legacy Module is now working. The hardware loop has been updated for better performance.
molyswu / Hand Detectionusing Neural Networks (SSD) on Tensorflow. This repo documents steps and scripts used to train a hand detector using Tensorflow (Object Detection API). As with any DNN based task, the most expensive (and riskiest) part of the process has to do with finding or creating the right (annotated) dataset. I was interested mainly in detecting hands on a table (egocentric view point). I experimented first with the [Oxford Hands Dataset](http://www.robots.ox.ac.uk/~vgg/data/hands/) (the results were not good). I then tried the [Egohands Dataset](http://vision.soic.indiana.edu/projects/egohands/) which was a much better fit to my requirements. The goal of this repo/post is to demonstrate how neural networks can be applied to the (hard) problem of tracking hands (egocentric and other views). Better still, provide code that can be adapted to other uses cases. If you use this tutorial or models in your research or project, please cite [this](#citing-this-tutorial). Here is the detector in action. <img src="images/hand1.gif" width="33.3%"><img src="images/hand2.gif" width="33.3%"><img src="images/hand3.gif" width="33.3%"> Realtime detection on video stream from a webcam . <img src="images/chess1.gif" width="33.3%"><img src="images/chess2.gif" width="33.3%"><img src="images/chess3.gif" width="33.3%"> Detection on a Youtube video. Both examples above were run on a macbook pro **CPU** (i7, 2.5GHz, 16GB). Some fps numbers are: | FPS | Image Size | Device| Comments| | ------------- | ------------- | ------------- | ------------- | | 21 | 320 * 240 | Macbook pro (i7, 2.5GHz, 16GB) | Run without visualizing results| | 16 | 320 * 240 | Macbook pro (i7, 2.5GHz, 16GB) | Run while visualizing results (image above) | | 11 | 640 * 480 | Macbook pro (i7, 2.5GHz, 16GB) | Run while visualizing results (image above) | > Note: The code in this repo is written and tested with Tensorflow `1.4.0-rc0`. Using a different version may result in [some errors](https://github.com/tensorflow/models/issues/1581). You may need to [generate your own frozen model](https://pythonprogramming.net/testing-custom-object-detector-tensorflow-object-detection-api-tutorial/?completed=/training-custom-objects-tensorflow-object-detection-api-tutorial/) graph using the [model checkpoints](model-checkpoint) in the repo to fit your TF version. **Content of this document** - Motivation - Why Track/Detect hands with Neural Networks - Data preparation and network training in Tensorflow (Dataset, Import, Training) - Training the hand detection Model - Using the Detector to Detect/Track hands - Thoughts on Optimizations. > P.S if you are using or have used the models provided here, feel free to reach out on twitter ([@vykthur](https://twitter.com/vykthur)) and share your work! ## Motivation - Why Track/Detect hands with Neural Networks? There are several existing approaches to tracking hands in the computer vision domain. Incidentally, many of these approaches are rule based (e.g extracting background based on texture and boundary features, distinguishing between hands and background using color histograms and HOG classifiers,) making them not very robust. For example, these algorithms might get confused if the background is unusual or in situations where sharp changes in lighting conditions cause sharp changes in skin color or the tracked object becomes occluded.(see [here for a review](https://www.cse.unr.edu/~bebis/handposerev.pdf) paper on hand pose estimation from the HCI perspective) With sufficiently large datasets, neural networks provide opportunity to train models that perform well and address challenges of existing object tracking/detection algorithms - varied/poor lighting, noisy environments, diverse viewpoints and even occlusion. The main drawbacks to usage for real-time tracking/detection is that they can be complex, are relatively slow compared to tracking-only algorithms and it can be quite expensive to assemble a good dataset. But things are changing with advances in fast neural networks. Furthermore, this entire area of work has been made more approachable by deep learning frameworks (such as the tensorflow object detection api) that simplify the process of training a model for custom object detection. More importantly, the advent of fast neural network models like ssd, faster r-cnn, rfcn (see [here](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md#coco-trained-models-coco-models) ) etc make neural networks an attractive candidate for real-time detection (and tracking) applications. Hopefully, this repo demonstrates this. > If you are not interested in the process of training the detector, you can skip straight to applying the [pretrained model I provide in detecting hands](#detecting-hands). Training a model is a multi-stage process (assembling dataset, cleaning, splitting into training/test partitions and generating an inference graph). While I lightly touch on the details of these parts, there are a few other tutorials cover training a custom object detector using the tensorflow object detection api in more detail[ see [here](https://pythonprogramming.net/training-custom-objects-tensorflow-object-detection-api-tutorial/) and [here](https://towardsdatascience.com/how-to-train-your-own-object-detector-with-tensorflows-object-detector-api-bec72ecfe1d9) ]. I recommend you walk through those if interested in training a custom object detector from scratch. ## Data preparation and network training in Tensorflow (Dataset, Import, Training) **The Egohands Dataset** The hand detector model is built using data from the [Egohands Dataset](http://vision.soic.indiana.edu/projects/egohands/) dataset. This dataset works well for several reasons. It contains high quality, pixel level annotations (>15000 ground truth labels) where hands are located across 4800 images. All images are captured from an egocentric view (Google glass) across 48 different environments (indoor, outdoor) and activities (playing cards, chess, jenga, solving puzzles etc). <img src="images/egohandstrain.jpg" width="100%"> If you will be using the Egohands dataset, you can cite them as follows: > Bambach, Sven, et al. "Lending a hand: Detecting hands and recognizing activities in complex egocentric interactions." Proceedings of the IEEE International Conference on Computer Vision. 2015. The Egohands dataset (zip file with labelled data) contains 48 folders of locations where video data was collected (100 images per folder). ``` -- LOCATION_X -- frame_1.jpg -- frame_2.jpg ... -- frame_100.jpg -- polygons.mat // contains annotations for all 100 images in current folder -- LOCATION_Y -- frame_1.jpg -- frame_2.jpg ... -- frame_100.jpg -- polygons.mat // contains annotations for all 100 images in current folder ``` **Converting data to Tensorflow Format** Some initial work needs to be done to the Egohands dataset to transform it into the format (`tfrecord`) which Tensorflow needs to train a model. This repo contains `egohands_dataset_clean.py` a script that will help you generate these csv files. - Downloads the egohands datasets - Renames all files to include their directory names to ensure each filename is unique - Splits the dataset into train (80%), test (10%) and eval (10%) folders. - Reads in `polygons.mat` for each folder, generates bounding boxes and visualizes them to ensure correctness (see image above). - Once the script is done running, you should have an images folder containing three folders - train, test and eval. Each of these folders should also contain a csv label document each - `train_labels.csv`, `test_labels.csv` that can be used to generate `tfrecords` Note: While the egohands dataset provides four separate labels for hands (own left, own right, other left, and other right), for my purpose, I am only interested in the general `hand` class and label all training data as `hand`. You can modify the data prep script to generate `tfrecords` that support 4 labels. Next: convert your dataset + csv files to tfrecords. A helpful guide on this can be found [here](https://pythonprogramming.net/creating-tfrecord-files-tensorflow-object-detection-api-tutorial/).For each folder, you should be able to generate `train.record`, `test.record` required in the training process. ## Training the hand detection Model Now that the dataset has been assembled (and your tfrecords), the next task is to train a model based on this. With neural networks, it is possible to use a process called [transfer learning](https://www.tensorflow.org/tutorials/image_retraining) to shorten the amount of time needed to train the entire model. This means we can take an existing model (that has been trained well on a related domain (here image classification) and retrain its final layer(s) to detect hands for us. Sweet!. Given that neural networks sometimes have thousands or millions of parameters that can take weeks or months to train, transfer learning helps shorten training time to possibly hours. Tensorflow does offer a few models (in the tensorflow [model zoo](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md#coco-trained-models-coco-models)) and I chose to use the `ssd_mobilenet_v1_coco` model as my start point given it is currently (one of) the fastest models (read the SSD research [paper here](https://arxiv.org/pdf/1512.02325.pdf)). The training process can be done locally on your CPU machine which may take a while or better on a (cloud) GPU machine (which is what I did). For reference, training on my macbook pro (tensorflow compiled from source to take advantage of the mac's cpu architecture) the maximum speed I got was 5 seconds per step as opposed to the ~0.5 seconds per step I got with a GPU. For reference it would take about 12 days to run 200k steps on my mac (i7, 2.5GHz, 16GB) compared to ~5hrs on a GPU. > **Training on your own images**: Please use the [guide provided by Harrison from pythonprogramming](https://pythonprogramming.net/training-custom-objects-tensorflow-object-detection-api-tutorial/) on how to generate tfrecords given your label csv files and your images. The guide also covers how to start the training process if training locally. [see [here] (https://pythonprogramming.net/training-custom-objects-tensorflow-object-detection-api-tutorial/)]. If training in the cloud using a service like GCP, see the [guide here](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/running_on_cloud.md). As the training process progresses, the expectation is that total loss (errors) gets reduced to its possible minimum (about a value of 1 or thereabout). By observing the tensorboard graphs for total loss(see image below), it should be possible to get an idea of when the training process is complete (total loss does not decrease with further iterations/steps). I ran my training job for 200k steps (took about 5 hours) and stopped at a total Loss (errors) value of 2.575.(In retrospect, I could have stopped the training at about 50k steps and gotten a similar total loss value). With tensorflow, you can also run an evaluation concurrently that assesses your model to see how well it performs on the test data. A commonly used metric for performance is mean average precision (mAP) which is single number used to summarize the area under the precision-recall curve. mAP is a measure of how well the model generates a bounding box that has at least a 50% overlap with the ground truth bounding box in our test dataset. For the hand detector trained here, the mAP value was **0.9686@0.5IOU**. mAP values range from 0-1, the higher the better. <img src="images/accuracy.jpg" width="100%"> Once training is completed, the trained inference graph (`frozen_inference_graph.pb`) is then exported (see the earlier referenced guides for how to do this) and saved in the `hand_inference_graph` folder. Now its time to do some interesting detection. ## Using the Detector to Detect/Track hands If you have not done this yet, please following the guide on installing [Tensorflow and the Tensorflow object detection api](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/installation.md). This will walk you through setting up the tensorflow framework, cloning the tensorflow github repo and a guide on - Load the `frozen_inference_graph.pb` trained on the hands dataset as well as the corresponding label map. In this repo, this is done in the `utils/detector_utils.py` script by the `load_inference_graph` method. ```python detection_graph = tf.Graph() with detection_graph.as_default(): od_graph_def = tf.GraphDef() with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid: serialized_graph = fid.read() od_graph_def.ParseFromString(serialized_graph) tf.import_graph_def(od_graph_def, name='') sess = tf.Session(graph=detection_graph) print("> ====== Hand Inference graph loaded.") ``` - Detect hands. In this repo, this is done in the `utils/detector_utils.py` script by the `detect_objects` method. ```python (boxes, scores, classes, num) = sess.run( [detection_boxes, detection_scores, detection_classes, num_detections], feed_dict={image_tensor: image_np_expanded}) ``` - Visualize detected bounding detection_boxes. In this repo, this is done in the `utils/detector_utils.py` script by the `draw_box_on_image` method. This repo contains two scripts that tie all these steps together. - detect_multi_threaded.py : A threaded implementation for reading camera video input detection and detecting. Takes a set of command line flags to set parameters such as `--display` (visualize detections), image parameters `--width` and `--height`, videe `--source` (0 for camera) etc. - detect_single_threaded.py : Same as above, but single threaded. This script works for video files by setting the video source parameter videe `--source` (path to a video file). ```cmd # load and run detection on video at path "videos/chess.mov" python detect_single_threaded.py --source videos/chess.mov ``` > Update: If you do have errors loading the frozen inference graph in this repo, feel free to generate a new graph that fits your TF version from the model-checkpoint in this repo. Use the [export_inference_graph.py](https://github.com/tensorflow/models/blob/master/research/object_detection/export_inference_graph.py) script provided in the tensorflow object detection api repo. More guidance on this [here](https://pythonprogramming.net/testing-custom-object-detector-tensorflow-object-detection-api-tutorial/?completed=/training-custom-objects-tensorflow-object-detection-api-tutorial/). ## Thoughts on Optimization. A few things that led to noticeable performance increases. - Threading: Turns out that reading images from a webcam is a heavy I/O event and if run on the main application thread can slow down the program. I implemented some good ideas from [Adrian Rosebuck](https://www.pyimagesearch.com/2017/02/06/faster-video-file-fps-with-cv2-videocapture-and-opencv/) on parrallelizing image capture across multiple worker threads. This mostly led to an FPS increase of about 5 points. - For those new to Opencv, images from the `cv2.read()` method return images in [BGR format](https://www.learnopencv.com/why-does-opencv-use-bgr-color-format/). Ensure you convert to RGB before detection (accuracy will be much reduced if you dont). ```python cv2.cvtColor(image_np, cv2.COLOR_BGR2RGB) ``` - Keeping your input image small will increase fps without any significant accuracy drop.(I used about 320 x 240 compared to the 1280 x 720 which my webcam provides). - Model Quantization. Moving from the current 32 bit to 8 bit can achieve up to 4x reduction in memory required to load and store models. One way to further speed up this model is to explore the use of [8-bit fixed point quantization](https://heartbeat.fritz.ai/8-bit-quantization-and-tensorflow-lite-speeding-up-mobile-inference-with-low-precision-a882dfcafbbd). Performance can also be increased by a clever combination of tracking algorithms with the already decent detection and this is something I am still experimenting with. Have ideas for optimizing better, please share! <img src="images/general.jpg" width="100%"> Note: The detector does reflect some limitations associated with the training set. This includes non-egocentric viewpoints, very noisy backgrounds (e.g in a sea of hands) and sometimes skin tone. There is opportunity to improve these with additional data. ## Integrating Multiple DNNs. One way to make things more interesting is to integrate our new knowledge of where "hands" are with other detectors trained to recognize other objects. Unfortunately, while our hand detector can in fact detect hands, it cannot detect other objects (a factor or how it is trained). To create a detector that classifies multiple different objects would mean a long involved process of assembling datasets for each class and a lengthy training process. > Given the above, a potential strategy is to explore structures that allow us **efficiently** interleave output form multiple pretrained models for various object classes and have them detect multiple objects on a single image. An example of this is with my primary use case where I am interested in understanding the position of objects on a table with respect to hands on same table. I am currently doing some work on a threaded application that loads multiple detectors and outputs bounding boxes on a single image. More on this soon.
Aastha2104 / Parkinson Disease PredictionIntroduction Parkinson’s Disease is the second most prevalent neurodegenerative disorder after Alzheimer’s, affecting more than 10 million people worldwide. Parkinson’s is characterized primarily by the deterioration of motor and cognitive ability. There is no single test which can be administered for diagnosis. Instead, doctors must perform a careful clinical analysis of the patient’s medical history. Unfortunately, this method of diagnosis is highly inaccurate. A study from the National Institute of Neurological Disorders finds that early diagnosis (having symptoms for 5 years or less) is only 53% accurate. This is not much better than random guessing, but an early diagnosis is critical to effective treatment. Because of these difficulties, I investigate a machine learning approach to accurately diagnose Parkinson’s, using a dataset of various speech features (a non-invasive yet characteristic tool) from the University of Oxford. Why speech features? Speech is very predictive and characteristic of Parkinson’s disease; almost every Parkinson’s patient experiences severe vocal degradation (inability to produce sustained phonations, tremor, hoarseness), so it makes sense to use voice to diagnose the disease. Voice analysis gives the added benefit of being non-invasive, inexpensive, and very easy to extract clinically. Background Parkinson's Disease Parkinson’s is a progressive neurodegenerative condition resulting from the death of the dopamine containing cells of the substantia nigra (which plays an important role in movement). Symptoms include: “frozen” facial features, bradykinesia (slowness of movement), akinesia (impairment of voluntary movement), tremor, and voice impairment. Typically, by the time the disease is diagnosed, 60% of nigrostriatal neurons have degenerated, and 80% of striatal dopamine have been depleted. Performance Metrics TP = true positive, FP = false positive, TN = true negative, FN = false negative Accuracy: (TP+TN)/(P+N) Matthews Correlation Coefficient: 1=perfect, 0=random, -1=completely inaccurate Algorithms Employed Logistic Regression (LR): Uses the sigmoid logistic equation with weights (coefficient values) and biases (constants) to model the probability of a certain class for binary classification. An output of 1 represents one class, and an output of 0 represents the other. Training the model will learn the optimal weights and biases. Linear Discriminant Analysis (LDA): Assumes that the data is Gaussian and each feature has the same variance. LDA estimates the mean and variance for each class from the training data, and then uses properties of statistics (Bayes theorem , Gaussian distribution, etc) to compute the probability of a particular instance belonging to a given class. The class with the largest probability is the prediction. k Nearest Neighbors (KNN): Makes predictions about the validation set using the entire training set. KNN makes a prediction about a new instance by searching through the entire set to find the k “closest” instances. “Closeness” is determined using a proximity measurement (Euclidean) across all features. The class that the majority of the k closest instances belong to is the class that the model predicts the new instance to be. Decision Tree (DT): Represented by a binary tree, where each root node represents an input variable and a split point, and each leaf node contains an output used to make a prediction. Neural Network (NN): Models the way the human brain makes decisions. Each neuron takes in 1+ inputs, and then uses an activation function to process the input with weights and biases to produce an output. Neurons can be arranged into layers, and multiple layers can form a network to model complex decisions. Training the network involves using the training instances to optimize the weights and biases. Naive Bayes (NB): Simplifies the calculation of probabilities by assuming that all features are independent of one another (a strong but effective assumption). Employs Bayes Theorem to calculate the probabilities that the instance to be predicted is in each class, then finds the class with the highest probability. Gradient Boost (GB): Generally used when seeking a model with very high predictive performance. Used to reduce bias and variance (“error”) by combining multiple “weak learners” (not very good models) to create a “strong learner” (high performance model). Involves 3 elements: a loss function (error function) to be optimized, a weak learner (decision tree) to make predictions, and an additive model to add trees to minimize the loss function. Gradient descent is used to minimize error after adding each tree (one by one). Engineering Goal Produce a machine learning model to diagnose Parkinson’s disease given various features of a patient’s speech with at least 90% accuracy and/or a Matthews Correlation Coefficient of at least 0.9. Compare various algorithms and parameters to determine the best model for predicting Parkinson’s. Dataset Description Source: the University of Oxford 195 instances (147 subjects with Parkinson’s, 48 without Parkinson’s) 22 features (elements that are possibly characteristic of Parkinson’s, such as frequency, pitch, amplitude / period of the sound wave) 1 label (1 for Parkinson’s, 0 for no Parkinson’s) Project Pipeline pipeline Summary of Procedure Split the Oxford Parkinson’s Dataset into two parts: one for training, one for validation (evaluate how well the model performs) Train each of the following algorithms with the training set: Logistic Regression, Linear Discriminant Analysis, k Nearest Neighbors, Decision Tree, Neural Network, Naive Bayes, Gradient Boost Evaluate results using the validation set Repeat for the following training set to validation set splits: 80% training / 20% validation, 75% / 25%, and 70% / 30% Repeat for a rescaled version of the dataset (scale all the numbers in the dataset to a range from 0 to 1: this helps to reduce the effect of outliers) Conduct 5 trials and average the results Data a_o a_r m_o m_r Data Analysis In general, the models tended to perform the best (both in terms of accuracy and Matthews Correlation Coefficient) on the rescaled dataset with a 75-25 train-test split. The two highest performing algorithms, k Nearest Neighbors and the Neural Network, both achieved an accuracy of 98%. The NN achieved a MCC of 0.96, while KNN achieved a MCC of 0.94. These figures outperform most existing literature and significantly outperform current methods of diagnosis. Conclusion and Significance These robust results suggest that a machine learning approach can indeed be implemented to significantly improve diagnosis methods of Parkinson’s disease. Given the necessity of early diagnosis for effective treatment, my machine learning models provide a very promising alternative to the current, rather ineffective method of diagnosis. Current methods of early diagnosis are only 53% accurate, while my machine learning model produces 98% accuracy. This 45% increase is critical because an accurate, early diagnosis is needed to effectively treat the disease. Typically, by the time the disease is diagnosed, 60% of nigrostriatal neurons have degenerated, and 80% of striatal dopamine have been depleted. With an earlier diagnosis, much of this degradation could have been slowed or treated. My results are very significant because Parkinson’s affects over 10 million people worldwide who could benefit greatly from an early, accurate diagnosis. Not only is my machine learning approach more accurate in terms of diagnostic accuracy, it is also more scalable, less expensive, and therefore more accessible to people who might not have access to established medical facilities and professionals. The diagnosis is also much simpler, requiring only a 10-15 second voice recording and producing an immediate diagnosis. Future Research Given more time and resources, I would investigate the following: Create a mobile application which would allow the user to record his/her voice, extract the necessary vocal features, and feed it into my machine learning model to diagnose Parkinson’s. Use larger datasets in conjunction with the University of Oxford dataset. Tune and improve my models even further to achieve even better results. Investigate different structures and types of neural networks. Construct a novel algorithm specifically suited for the prediction of Parkinson’s. Generalize my findings and algorithms for all types of dementia disorders, such as Alzheimer’s. References Bind, Shubham. "A Survey of Machine Learning Based Approaches for Parkinson Disease Prediction." International Journal of Computer Science and Information Technologies 6 (2015): n. pag. International Journal of Computer Science and Information Technologies. 2015. Web. 8 Mar. 2017. Brooks, Megan. "Diagnosing Parkinson's Disease Still Challenging." Medscape Medical News. National Institute of Neurological Disorders, 31 July 2014. Web. 20 Mar. 2017. Exploiting Nonlinear Recurrence and Fractal Scaling Properties for Voice Disorder Detection', Little MA, McSharry PE, Roberts SJ, Costello DAE, Moroz IM. BioMedical Engineering OnLine 2007, 6:23 (26 June 2007) Hashmi, Sumaiya F. "A Machine Learning Approach to Diagnosis of Parkinson’s Disease."Claremont Colleges Scholarship. Claremont College, 2013. Web. 10 Mar. 2017. Karplus, Abraham. "Machine Learning Algorithms for Cancer Diagnosis." Machine Learning Algorithms for Cancer Diagnosis (n.d.): n. pag. Mar. 2012. Web. 20 Mar. 2017. Little, Max. "Parkinsons Data Set." UCI Machine Learning Repository. University of Oxford, 26 June 2008. Web. 20 Feb. 2017. Ozcift, Akin, and Arif Gulten. "Classifier Ensemble Construction with Rotation Forest to Improve Medical Diagnosis Performance of Machine Learning Algorithms." Computer Methods and Programs in Biomedicine 104.3 (2011): 443-51. Semantic Scholar. 2011. Web. 15 Mar. 2017. "Parkinson’s Disease Dementia." UCI MIND. N.p., 19 Oct. 2015. Web. 17 Feb. 2017. Salvatore, C., A. Cerasa, I. Castiglioni, F. Gallivanone, A. Augimeri, M. Lopez, G. Arabia, M. Morelli, M.c. Gilardi, and A. Quattrone. "Machine Learning on Brain MRI Data for Differential Diagnosis of Parkinson's Disease and Progressive Supranuclear Palsy."Journal of Neuroscience Methods 222 (2014): 230-37. 2014. Web. 18 Mar. 2017. Shahbakhi, Mohammad, Danial Taheri Far, and Ehsan Tahami. "Speech Analysis for Diagnosis of Parkinson’s Disease Using Genetic Algorithm and Support Vector Machine."Journal of Biomedical Science and Engineering 07.04 (2014): 147-56. Scientific Research. July 2014. Web. 2 Mar. 2017. "Speech and Communication." Speech and Communication. Parkinson's Disease Foundation, n.d. Web. 22 Mar. 2017. Sriram, Tarigoppula V. S., M. Venkateswara Rao, G. V. Satya Narayana, and D. S. V. G. K. Kaladhar. "Diagnosis of Parkinson Disease Using Machine Learning and Data Mining Systems from Voice Dataset." SpringerLink. Springer, Cham, 01 Jan. 1970. Web. 17 Mar. 2017.
USTCPCS / CVPR2018 AttentionContext Encoding for Semantic Segmentation MegaDepth: Learning Single-View Depth Prediction from Internet Photos LiteFlowNet: A Lightweight Convolutional Neural Network for Optical Flow Estimation PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume On the Robustness of Semantic Segmentation Models to Adversarial Attacks SPLATNet: Sparse Lattice Networks for Point Cloud Processing Left-Right Comparative Recurrent Model for Stereo Matching Enhancing the Spatial Resolution of Stereo Images using a Parallax Prior Unsupervised CCA Discovering Point Lights with Intensity Distance Fields CBMV: A Coalesced Bidirectional Matching Volume for Disparity Estimation Learning a Discriminative Feature Network for Semantic Segmentation Revisiting Dilated Convolution: A Simple Approach for Weakly- and Semi- Supervised Semantic Segmentation Unsupervised Deep Generative Adversarial Hashing Network Monocular Relative Depth Perception with Web Stereo Data Supervision Single Image Reflection Separation with Perceptual Losses Zoom and Learn: Generalizing Deep Stereo Matching to Novel Domains EPINET: A Fully-Convolutional Neural Network for Light Field Depth Estimation by Using Epipolar Geometry FoldingNet: Interpretable Unsupervised Learning on 3D Point Clouds Decorrelated Batch Normalization Unsupervised Learning of Depth and Egomotion from Monocular Video Using 3D Geometric Constraints PU-Net: Point Cloud Upsampling Network Real-Time Monocular Depth Estimation using Synthetic Data with Domain Adaptation via Image Style Transfer Tell Me Where To Look: Guided Attention Inference Network Residual Dense Network for Image Super-Resolution Reflection Removal for Large-Scale 3D Point Clouds PlaneNet: Piece-wise Planar Reconstruction from a Single RGB Image Fully Convolutional Adaptation Networks for Semantic Segmentation CRRN: Multi-Scale Guided Concurrent Reflection Removal Network DenseASPP: Densely Connected Networks for Semantic Segmentation SGAN: An Alternative Training of Generative Adversarial Networks Multi-Agent Diverse Generative Adversarial Networks Robust Depth Estimation from Auto Bracketed Images AdaDepth: Unsupervised Content Congruent Adaptation for Depth Estimation DeepMVS: Learning Multi-View Stereopsis GeoNet: Unsupervised Learning of Dense Depth, Optical Flow and Camera Pose GeoNet: Geometric Neural Network for Joint Depth and Surface Normal Estimation Single-Image Depth Estimation Based on Fourier Domain Analysis Single View Stereo Matching Pyramid Stereo Matching Network A Unifying Contrast Maximization Framework for Event Cameras, with Applications to Motion, Depth, and Optical Flow Estimation Image Correction via Deep Reciprocating HDR Transformation Occlusion Aware Unsupervised Learning of Optical Flow PAD-Net: Multi-Tasks Guided Prediciton-and-Distillation Network for Simultaneous Depth Estimation and Scene Parsing Surface Networks Structured Attention Guided Convolutional Neural Fields for Monocular Depth Estimation TextureGAN: Controlling Deep Image Synthesis with Texture Patches Aperture Supervision for Monocular Depth Estimation Two-Stream Convolutional Networks for Dynamic Texture Synthesis Unsupervised Learning of Single View Depth Estimation and Visual Odometry with Deep Feature Reconstruction Left/Right Asymmetric Layer Skippable Networks Learning to See in the Dark
jettbrains / L W3C Strategic Highlights September 2019 This report was prepared for the September 2019 W3C Advisory Committee Meeting (W3C Member link). See the accompanying W3C Fact Sheet — September 2019. For the previous edition, see the April 2019 W3C Strategic Highlights. For future editions of this report, please consult the latest version. A Chinese translation is available. ☰ Contents Introduction Future Web Standards Meeting Industry Needs Web Payments Digital Publishing Media and Entertainment Web & Telecommunications Real-Time Communications (WebRTC) Web & Networks Automotive Web of Things Strengthening the Core of the Web HTML CSS Fonts SVG Audio Performance Web Performance WebAssembly Testing Browser Testing and Tools WebPlatform Tests Web of Data Web for All Security, Privacy, Identity Internationalization (i18n) Web Accessibility Outreach to the world W3C Developer Relations W3C Training Translations W3C Liaisons Introduction This report highlights recent work of enhancement of the existing landscape of the Web platform and innovation for the growth and strength of the Web. 33 working groups and a dozen interest groups enable W3C to pursue its mission through the creation of Web standards, guidelines, and supporting materials. We track the tremendous work done across the Consortium through homogeneous work-spaces in Github which enables better monitoring and management. We are in the middle of a period where we are chartering numerous working groups which demonstrate the rapid degree of change for the Web platform: After 4 years, we are nearly ready to publish a Payment Request API Proposed Recommendation and we need to soon charter follow-on work. In the last year we chartered the Web Payment Security Interest Group. In the last year we chartered the Web Media Working Group with 7 specifications for next generation Media support on the Web. We have Accessibility Guidelines under W3C Member review which includes Silver, a new approach. We have just launched the Decentralized Identifier Working Group which has tremendous potential because Decentralized Identifier (DID) is an identifier that is globally unique, resolveable with high availability, and cryptographically verifiable. We have Privacy IG (PING) under W3C Member review which strengthens our focus on the tradeoff between privacy and function. We have a new CSS charter under W3C Member review which maps the group's work for the next three years. In this period, W3C and the WHATWG have succesfully completed the negotiation of a Memorandum of Understanding rooted in the mutual belief that that having two distinct specifications claiming to be normative is generally harmful for the Web community. The MOU, signed last May, describes how the two organizations are to collaborate on the development of a single authoritative version of the HTML and DOM specifications. W3C subsequently rechartered the HTML Working Group to assist the W3C community in raising issues and proposing solutions for the HTML and DOM specifications, and for the production of W3C Recommendations from WHATWG Review Drafts. As the Web evolves continuously, some groups are looking for ways for specifications to do so as well. So-called "evergreen recommendations" or "living standards" aim to track continuous development (and maintenance) of features, on a feature-by-feature basis, while getting review and patent commitments. We see the maturation and further development of an incredible number of new technologies coming to the Web. Continued progress in many areas demonstrates the vitality of the W3C and the Web community, as the rest of the report illustrates. Future Web Standards W3C has a variety of mechanisms for listening to what the community thinks could become good future Web standards. These include discussions with the Membership, discussions with other standards bodies, the activities of thousands of participants in over 300 community groups, and W3C Workshops. There are lots of good ideas. The W3C strategy team has been identifying promising topics and invites public participation. Future, recent and under consideration Workshops include: Inclusive XR (5-6 November 2019, Seattle, WA, USA) to explore existing and future approaches on making Virtual and Augmented Reality experiences more inclusive, including to people with disabilities; W3C Workshop on Data Models for Transportation (12-13 September 2019, Palo Alto, CA, USA) W3C Workshop on Web Games (27-28 June 2019, Redmond, WA, USA), view report Second W3C Workshop on the Web of Things (3-5 June 2019, Munich, Germany) W3C Workshop on Web Standardization for Graph Data; Creating Bridges: RDF, Property Graph and SQL (4-6 March 2019, Berlin, Germany), view report Web & Machine Learning. The Strategy Funnel documents the staff's exploration of potential new work at various phases: Exploration and Investigation, Incubation and Evaluation, and eventually to the chartering of a new standards group. The Funnel view is a GitHub Project where new area are issues represented by “cards” which move through the columns, usually from left to right. Most cards start in Exploration and move towards Chartering, or move out of the funnel. Public input is welcome at any stage but particularly once Incubation has begun. This helps W3C identify work that is sufficiently incubated to warrant standardization, to review the ecosystem around the work and indicate interest in participating in its standardization, and then to draft a charter that reflects an appropriate scope. Ongoing feedback can speed up the overall standardization process. Since the previous highlights document, W3C has chartered a number of groups, and started discussion on many more: Newly Chartered or Rechartered Web Application Security WG (03-Apr) Web Payment Security IG (17-Apr) Patent and Standards IG (24-Apr) Web Applications WG (14-May) Web & Networks IG (16-May) Media WG (23-May) Media and Entertainment IG (06-Jun) HTML WG (06-Jun) Decentralized Identifier WG (05-Sep) Extended Privacy IG (PING) (30-Sep) Verifiable Claims WG (30-Sep) Service Workers WG (31-Dec) Dataset Exchange WG (31-Dec) Web of Things Working Group (31-Dec) Web Audio Working Group (31-Dec) Proposed charters / Advance Notice Accessibility Guidelines WG Privacy IG (PING) RDF Literal Direction WG Timed Text WG CSS WG Web Authentication WG Closed Internationalization Tag Set IG Meeting Industry Needs Web Payments All Web Payments specifications W3C's payments standards enable a streamlined checkout experience, enabling a consistent user experience across the Web with lower front end development costs for merchants. Users can store and reuse information and more quickly and accurately complete online transactions. The Web Payments Working Group has republished Payment Request API as a Candidate Recommendation, aiming to publish a Proposed Recommendation in the Fall 2019, and is discussing use cases and features for Payment Request after publication of the 1.0 Recommendation. Browser vendors have been finalizing implementation of features added in the past year (view the implementation report). As work continues on the Payment Handler API and its implementation (currently in Chrome and Edge Canary), one focus in 2019 is to increase adoption in other browsers. Recently, Mastercard demonstrated the use of Payment Request API to carry out EMVCo's Secure Remote Commerce (SRC) protocol whose payment method definition is being developed with active participation by Visa, Mastercard, American Express, and Discover. Payment method availability is a key factor in merchant considerations about adopting Payment Request API. The ability to get uniform adoption of a new payment method such as Secure Remote Commerce (SRC) also depends on the availability of the Payment Handler API in browsers, or of proprietary alternatives. Web Monetization, which the Web Payments Working Group will discuss again at its face-to-face meeting in September, can be used to enable micropayments as an alternative revenue stream to advertising. Since the beginning of 2019, Amazon, Brave Software, JCB, Certus Cybersecurity Solutions and Netflix have joined the Web Payments Working Group. In April, W3C launched the Web Payment Security Group to enable W3C, EMVCo, and the FIDO Alliance to collaborate on a vision for Web payment security and interoperability. Participants will define areas of collaboration and identify gaps between existing technical specifications in order to increase compatibility among different technologies, such as: How do SRC, FIDO, and Payment Request relate? The Payment Services Directive 2 (PSD2) regulations in Europe are scheduled to take effect in September 2019. What is the role of EMVCo, W3C, and FIDO technologies, and what is the current state of readiness for the deadline? How can we improve privacy on the Web at the same time as we meet industry requirements regarding user identity? Digital Publishing All Digital Publishing specifications, Publication milestones The Web is the universal publishing platform. Publishing is increasingly impacted by the Web, and the Web increasingly impacts Publishing. Topic of particular interest to Publishing@W3C include typography and layout, accessibility, usability, portability, distribution, archiving, offline access, print on demand, and reliable cross referencing. And the diverse publishing community represented in the groups consist of the traditional "trade" publishers, ebook reading system manufacturers, but also publishers of audio book, scholarly journals or educational materials, library scientists or browser developers. The Publishing Working Group currently concentrates on Audiobooks which lack a comprehensive standard, thus incurring extra costs and time to publish in this booming market. Active development is ongoing on the future standard: Publication Manifest Audiobook profile for Web Publications Lightweight Packaging Format The BD Comics Manga Community Group, the Synchronized Multimedia for Publications Community Group, the Publishing Community Group and a future group on archival, are companions to the working group where specific work is developed and incubated. The Publishing Community Group is a recently launched incubation channel for Publishing@W3C. The goal of the group is to propose, document, and prototype features broadly related to: publications on the Web reading modes and systems and the user experience of publications The EPUB 3 Community Group has successfully completed the revision of EPUB 3.2. The Publishing Business Group fosters ongoing participation by members of the publishing industry and the overall ecosystem in the development of Web infrastructure to better support the needs of the industry. The Business Group serves as an additional conduit to the Publishing Working Group and several Community Groups for feedback between the publishing ecosystem and W3C. The Publishing BG has played a vital role in fostering and advancing the adoption and continued development of EPUB 3. In particular the BG provided critical support to the update of EPUBCheck to validate EPUB content to the new EPUB 3.2 specification. This resulted in the development, in conjunction with the EPUB3 Community Group, of a new generation of EPUBCheck, i.e., EPUBCheck 4.2 production-ready release. Media and Entertainment All Media specifications The Media and Entertainment vertical tracks media-related topics and features that create immersive experiences for end users. HTML5 brought standard audio and video elements to the Web. Standardization activities since then have aimed at turning the Web into a professional platform fully suitable for the delivery of media content and associated materials, enabling missing features to stream video content on the Web such as adaptive streaming and content protection. Together with Microsoft, Comcast, Netflix and Google, W3C received an Technology & Engineering Emmy Award in April 2019 for standardization of a full TV experience on the Web. Current goals are to: Reinforce core media technologies: Creation of the Media Working Group, to develop media-related specifications incubated in the WICG (e.g. Media Capabilities, Picture-in-picture, Media Session) and maintain maintain/evolve Media Source Extensions (MSE) and Encrypted Media Extensions (EME). Improve support for Media Timed Events: data cues incubation. Enhance color support (HDR, wide gamut), in scope of the CSS WG and in the Color on the Web CG. Reduce fragmentation: Continue annual releases of a common and testable baseline media devices, in scope of the Web Media APIs CG and in collaboration with the CTA WAVE Project. Maintain the Road-map of Media Technologies for the Web which highlights Web technologies that can be used to build media applications and services, as well as known gaps to enable additional use cases. Create the future: Discuss perspectives for Media and Entertainment for the Web. Bring the power of GPUs to the Web (graphics, machine learning, heavy processing), under incubation in the GPU for the Web CG. Transition to a Working Group is under discussion. Determine next steps after the successful W3C Workshop on Web Games of June 2019. View the report. Timed Text The Timed Text Working Group develops and maintains formats used for the representation of text synchronized with other timed media, like audio and video, and notably works on TTML, profiles of TTML, and WebVTT. Recent progress includes: A robust WebVTT implementation report poises the specification for publication as a proposed recommendation. Discussions around re-chartering, notably to add a TTML Profile for Audio Description deliverable to the scope of the group, and clarify that rendering of captions within XR content is also in scope. Immersive Web Hardware that enables Virtual Reality (VR) and Augmented Reality (AR) applications are now broadly available to consumers, offering an immersive computing platform with both new opportunities and challenges. The ability to interact directly with immersive hardware is critical to ensuring that the web is well equipped to operate as a first-class citizen in this environment. The Immersive Web Working Group has been stabilizing the WebXR Device API while the companion Immersive Web Community Group incubates the next series of features identified as key for the future of the Immersive Web. W3C plans a workshop focused on the needs and benefits at the intersection of VR & Accessibility (Inclusive XR), on 5-6 November 2019 in Seattle, WA, USA, to explore existing and future approaches on making Virtual and Augmented Reality experiences more inclusive. Web & Telecommunications The Web is the Open Platform for Mobile. Telecommunication service providers and network equipment providers have long been critical actors in the deployment of Web technologies. As the Web platform matures, it brings richer and richer capabilities to extend existing services to new users and devices, and propose new and innovative services. Real-Time Communications (WebRTC) All Real-Time Communications specifications WebRTC has reshaped the whole communication landscape by making any connected device a potential communication end-point, bringing audio and video communications anywhere, on any network, vastly expanding the ability of operators to reach their customers. WebRTC serves as the corner-stone of many online communication and collaboration services. The WebRTC Working Group aims to bringing WebRTC 1.0 (and companion specification Media Capture and Streams) to Recommendation by the end of 2019. Intense efforts are focused on testing (supported by a dedicated hackathon at IETF 104) and interoperability. The group is considering pushing features that have not gotten enough traction to separate modules or to a later minor revision of the spec. Beyond WebRTC 1.0, the WebRTC Working Group will focus its efforts on WebRTC NV which the group has started documenting by identifying use cases. Web & Networks Recently launched, in the wake of the May 2018 Web5G workshop, the Web & Networks Interest Group is chaired by representatives from AT&T, China Mobile and Intel, with a goal to explore solutions for web applications to achieve better performance and resource allocation, both on the device and network. The group's first efforts are around use cases, privacy & security requirements and liaisons. Automotive All Automotive specifications To create a rich application ecosystem for vehicles and other devices allowed to connect to the vehicle, the W3C Automotive Working Group is delivering a service specification to expose all common vehicle signals (engine temperature, fuel/charge level, range, tire pressure, speed, etc.) The Vehicle Information Service Specification (VISS), which is a Candidate Recommendation, is seeing more implementations across the industry. It provides the access method to a common data model for all the vehicle signals –presently encapsulating a thousand or so different data elements– and will be growing to accommodate the advances in automotive such as autonomous and driver assist technologies and electrification. The group is already working on a successor to VISS, leveraging the underlying data model and the VIWI submission from Volkswagen, for a more robust means of accessing vehicle signals information and the same paradigm for other automotive needs including location-based services, media, notifications and caching content. The Automotive and Web Platform Business Group acts as an incubator for prospective standards work. One of its task forces is using W3C VISS in performing data sampling and off-boarding the information to the cloud. Access to the wealth of information that W3C's auto signals standard exposes is of interest to regulators, urban planners, insurance companies, auto manufacturers, fleet managers and owners, service providers and others. In addition to components needed for data sampling and edge computing, capturing user and owner consent, information collection methods and handling of data are in scope. The upcoming W3C Workshop on Data Models for Transportation (September 2019) is expected to focus on the need of additional ontologies around transportation space. Web of Things All Web of Things specifications W3C's Web of Things work is designed to bridge disparate technology stacks to allow devices to work together and achieve scale, thus enabling the potential of the Internet of Things by eliminating fragmentation and fostering interoperability. Thing descriptions expressed in JSON-LD cover the behavior, interaction affordances, data schema, security configuration, and protocol bindings. The Web of Things complements existing IoT ecosystems to reduce the cost and risk for suppliers and consumers of applications that create value by combining multiple devices and information services. There are many sectors that will benefit, e.g. smart homes, smart cities, smart industry, smart agriculture, smart healthcare and many more. The Web of Things Working Group is finishing the initial Web of Things standards, with support from the Web of Things Interest Group: Web of Things Architecture Thing Descriptions Strengthening the Core of the Web HTML The HTML Working Group was chartered early June to assist the W3C community in raising issues and proposing solutions for the HTML and DOM specifications, and to produce W3C Recommendations from WHATWG Review Drafts. A few days before, W3C and the WHATWG signed a Memorandum of Understanding outlining the agreement to collaborate on the development of a single version of the HTML and DOM specifications. Issues and proposed solutions for HTML and DOM done via the newly rechartered HTML Working Group in the WHATWG repositories The HTML Working Group is targetting November 2019 to bring HTML and DOM to Candidate Recommendations. CSS All CSS specifications CSS is a critical part of the Open Web Platform. The CSS Working Group gathers requirements from two large groups of CSS users: the publishing industry and application developers. Within W3C, those groups are exemplified by the Publishing groups and the Web Platform Working Group. The former requires things like better pagination support and advanced font handling, the latter needs intelligent (and fast!) scrolling and animations. What we know as CSS is actually a collection of almost a hundred specifications, referred to as ‘modules’. The current state of CSS is defined by a snapshot, updated once a year. The group also publishes an index defining every term defined by CSS specifications. Fonts All Fonts specifications The Web Fonts Working Group develops specifications that allow the interoperable deployment of downloadable fonts on the Web, with a focus on Progressive Font Enrichment as well as maintenance of WOFF Recommendations. Recent and ongoing work includes: Early API experiments by Adobe and Monotype have demonstrated the feasibility of a font enrichment API, where a server delivers a font with minimal glyph repertoire and the client can query the full repertoire and request additional subsets on-the-fly. In other experiments, the Brotli compression used in WOFF 2 was extended to support shared dictionaries and patch update. Metrics to quantify improvement are a current hot discussion topic. The group will meet at ATypi 2019 in Japan, to gather requirements from the international typography community. The group will first produce a report summarizing the strengths and weaknesses of each prototype solution by Q2 2020. SVG All SVG specifications SVG is an important and widely-used part of the Open Web Platform. The SVG Working Group focuses on aligning the SVG 2.0 specification with browser implementations, having split the specification into a currently-implemented 2.0 and a forward-looking 2.1. Current activity is on stabilization, increased integration with the Open Web Platform, and test coverage analysis. The Working Group was rechartered in March 2019. A new work item concerns native (non-Web-browser) uses of SVG as a non-interactive, vector graphics format. Audio The Web Audio Working Group was extended to finish its work on the Web Audio API, expecting to publish it as a Recommendation by year end. The specification enables synthesizing audio in the browser. Audio operations are performed with audio nodes, which are linked together to form a modular audio routing graph. Multiple sources — with different types of channel layout — are supported. This modular design provides the flexibility to create complex audio functions with dynamic effects. The first version of Web Audio API is now feature complete and is implemented in all modern browsers. Work has started on the next version, and new features are being incubated in the Audio Community Group. Performance Web Performance All Web Performance specifications There are currently 18 specifications in development in the Web Performance Working Group aiming to provide methods to observe and improve aspects of application performance of user agent features and APIs. The W3C team is looking at related work incubated in the W3C GPU for the Web (WebGPU) Community Group which is poised to transition to a W3C Working Group. A preliminary draft charter is available. WebAssembly All WebAssembly specifications WebAssembly improves Web performance and power by being a virtual machine and execution environment enabling loaded pages to run native (compiled) code. It is deployed in Firefox, Edge, Safari and Chrome. The specification will soon reach Candidate Recommendation. WebAssembly enables near-native performance, optimized load time, and perhaps most importantly, a compilation target for existing code bases. While it has a small number of native types, much of the performance increase relative to Javascript derives from its use of consistent typing. WebAssembly leverages decades of optimization for compiled languages and the byte code is optimized for compactness and streaming (the web page starts executing while the rest of the code downloads). Network and API access all occurs through accompanying Javascript libraries -- the security model is identical to that of Javascript. Requirements gathering and language development occur in the Community Group while the Working Group manages test development, community review and progression of specifications on the Recommendation Track. Testing Browser testing plays a critical role in the growth of the Web by: Improving the reliability of Web technology definitions; Improving the quality of implementations of these technologies by helping vendors to detect bugs in their products; Improving the data available to Web developers on known bugs and deficiencies of Web technologies by publishing results of these tests. Browser Testing and Tools The Browser Testing and Tools Working Group is developing WebDriver version 2, having published last year the W3C Recommendation of WebDriver. WebDriver acts as a remote control interface that enables introspection and control of user agents, provides a platform- and language-neutral wire protocol as a way for out-of-process programs to remotely instruct the behavior of Web, and emulates the actions of a real person using the browser. WebPlatform Tests The WebPlatform Tests project now provides a mechanism which allows to fully automate tests that previously needed to be run manually: TestDriver. TestDriver enables sending trusted key and mouse events, sending complex series of trusted pointer and key interactions for things like in-content drag-and-drop or pinch zoom, and even file upload. Since 2014 W3C began work on this coordinated open-source effort to build a cross-browser test suite for the Web Platform, which WHATWG, and all major browsers adopted. Web of Data All Data specifications There have been several great success stories around the standardization of data on the web over the past year. Verifiable Claims seems to have significant uptake. It is also significant that the Distributed Identifier WG charter has received numerous favorable reviews, and was just recently launched. JSON-LD has been a major success with the large deployment on Web sites via schema.org. JSON-LD 1.1 completed technical work, about to transition to CR More than 25% of websites today include schema.org data in JSON-LD The Web of Things description is in CR since May, making use of JSON-LD Verifiable Credentials data model is in CR since July, also making use of JSON-LD Continued strong interest in decentralized identifiers Engagement from the TAG with reframing core documents, such as Ethical Web Principles, to include data on the web within their scope Data is increasingly important for all organizations, especially with the rise of IoT and Big Data. W3C has a mature and extensive suite of standards relating to data that were developed over two decades of experience, with plans for further work on making it easier for developers to work with graph data and knowledge graphs. Linked Data is about the use of URIs as names for things, the ability to dereference these URIs to get further information and to include links to other data. There are ever-increasing sources of open Linked Data on the Web, as well as data services that are restricted to the suppliers and consumers of those services. The digital transformation of industry is seeking to exploit advanced digital technologies. This will facilitate businesses to integrate horizontally along the supply and value chains, and vertically from the factory floor to the office floor. W3C is seeking to make it easier to support enterprise-wide data management and governance, reflecting the strategic importance of data to modern businesses. Traditional approaches to data have focused on tabular databases (SQL/RDBMS), Comma Separated Value (CSV) files, and data embedded in PDF documents and spreadsheets. We're now in midst of a major shift to graph data with nodes and labeled directed links between them. Graph data is: Faster than using SQL and associated JOIN operations More favorable to integrating data from heterogeneous sources Better suited to situations where the data model is evolving In the wake of the recent W3C Workshop on Graph Data we are in the process of launching a Graph Standardization Business Group to provide a business perspective with use cases and requirements, to coordinate technical standards work and liaisons with external organizations. Web for All Security, Privacy, Identity All Security specifications, all Privacy specifications Authentication on the Web As the WebAuthn Level 1 W3C Recommendation published last March is seeing wide implementation and adoption of strong cryptographic authentication, work is proceeding on Level 2. The open standard Web API gives native authentication technology built into native platforms, browsers, operating systems (including mobile) and hardware, offering protection against hacking, credential theft, phishing attacks, thus aiming to end the era of passwords as a security construct. You may read more in our March press release. Privacy An increasing number of W3C specifications are benefitting from Privacy and Security review; there are security and privacy aspects to every specification. Early review is essential. Working with the TAG, the Privacy Interest Group has updated the Self-Review Questionnaire: Security and Privacy. Other recent work of the group includes public blogging further to the exploration of anti-patterns in standards and permission prompts. Security The Web Application Security Working Group adopted Feature Policy, aiming to allow developers to selectively enable, disable, or modify the behavior of some of these browser features and APIs within their application; and Fetch Metadata, aiming to provide servers with enough information to make a priori decisions about whether or not to service a request based on the way it was made, and the context in which it will be used. The Web Payment Security Interest Group, launched last April, convenes members from W3C, EMVCo, and the FIDO Alliance to discuss cooperative work to enhance the security and interoperability of Web payments (read more about payments). Internationalization (i18n) All Internationalization specifications, educational articles related to Internationalization, spec developers checklist Only a quarter or so current Web users use English online and that proportion will continue to decrease as the Web reaches more and more communities of limited English proficiency. If the Web is to live up to the "World Wide" portion of its name, and for the Web to truly work for stakeholders all around the world engaging with content in various languages, it must support the needs of worldwide users as they engage with content in the various languages. The growth of epublishing also brings requirements for new features and improved typography on the Web. It is important to ensure the needs of local communities are captured. The W3C Internationalization Initiative was set up to increase in-house resources dedicated to accelerating progress in making the World Wide Web "worldwide" by gathering user requirements, supporting developers, and education & outreach. For an overview of current projects see the i18n radar. W3C's Internationalization efforts progressed on a number of fronts recently: Requirements: New African and European language groups will work on the gap analysis, errata and layout requirements. Gap analysis: Japanese, Devanagari, Bengali, Tamil, Lao, Khmer, Javanese, and Ethiopic updated in the gap-analysis documents. Layout requirements document: notable progress tracked in the Southeast Asian Task Force while work continues on Chinese layout requirements. Developer support: Spec reviews: the i18n WG continues active review of specifications of the WHATWG and other W3C Working Groups. Short review checklist: easy way to begin a self-review to help spec developers understand what aspects of their spec are likely to need attention for internationalization, and points them to more detailed checklists for the relevant topics. It also helps those reviewing specs for i18n issues. Strings on the Web: Language and Direction Metadata lays out issues and discusses potential solutions for passing information about language and direction with strings in JSON or other data formats. The document was rewritten for clarity, and expanded. The group is collaborating with the JSON-LD and Web Publishing groups to develop a plan for updating RDF, JSON-LD and related specifications to handle metadata for base direction of text (bidi). User-friendly test format: a new format was developed for Internationalization Test Suite tests, which displays helpful information about how the test works. This particularly useful because those tests are pointed to by educational materials and gap-analysis documents. Web Platform Tests: a large number of tests in the i18n test suite have been ported to the WPT repository, including: css-counter-styles, css-ruby, css-syntax, css-test, css-text-decor, css-writing-modes, and css-pseudo. Education & outreach: (for all educational materials, see the HTML & CSS Authoring Techniques) Web Accessibility All Accessibility specifications, WAI resources The Web Accessibility Initiative supports W3C's Web for All mission. Recent achievements include: Education and training: Inaccessibility of CAPTCHA updated to bring our analysis and recommendations up to date with CAPTCHA practice today, concluding two years of extensive work and invaluable input from the public (read more on the W3C Blog Learn why your web content and applications should be accessible. The Education and Outreach Working Group has completed revision and updating of the Business Case for Digital Accessibility. Accessibility guidelines: The Accessibility Guidelines Working Group has continued to update WCAG Techniques and Understanding WCAG 2.1; and published a Candidate Recommendation of Accessibility Conformance Testing Rules Format 1.0 to improve inter-rater reliability when evaluating conformance of web content to WCAG An updated charter is being developed to host work on "Silver", the next generation accessibility guidelines (WCAG 2.2) There are accessibility aspects to most specifications. Check your work with the FAST checklist. Outreach to the world W3C Developer Relations To foster the excellent feedback loop between Web Standards development and Web developers, and to grow participation from that diverse community, recent W3C Developer Relations activities include: @w3cdevs tracks the enormous amount of work happening across W3C W3C Track during the Web Conference 2019 in San Francisco Tech videos: W3C published the 2019 Web Games Workshop videos The 16 September 2019 Developer Meetup in Fukuoka, Japan, is open to all and will combine a set of technical demos prepared by W3C groups, and a series of talks on a selected set of W3C technologies and projects W3C is involved with Mozilla, Google, Samsung, Microsoft and Bocoup in the organization of ViewSource 2019 in Amsterdam (read more on the W3C Blog) W3C Training In partnership with EdX, W3C's MOOC training program, W3Cx offers a complete "Front-End Web Developer" (FEWD) professional certificate program that consists of a suite of five courses on the foundational languages that power the Web: HTML5, CSS and JavaScript. We count nearly 900K students from all over the world. Translations Many Web users rely on translations of documents developed at W3C whose official language is English. W3C is extremely grateful to the continuous efforts of its community in ensuring our various deliverables in general, and in our specifications in particular, are made available in other languages, for free, ensuring their exposure to a much more diverse set of readers. Last Spring we developed a more robust system, a new listing of translations of W3C specifications and updated the instructions on how to contribute to our translation efforts. W3C Liaisons Liaisons and coordination with numerous organizations and Standards Development Organizations (SDOs) is crucial for W3C to: make sure standards are interoperable coordinate our respective agenda in Internet governance: W3C participates in ICANN, GIPO, IGF, the I* organizations (ICANN, IETF, ISOC, IAB). ensure at the government liaison level that our standards work is officially recognized when important to our membership so that products based on them (often done by our members) are part of procurement orders. W3C has ARO/PAS status with ISO. W3C participates in the EU MSP and Rolling Plan on Standardization ensure the global set of Web and Internet standards form a compatible stack of technologies, at the technical and policy level (patent regime, fragmentation, use in policy making) promote Standards adoption equally by the industry, the public sector, and the public at large Coralie Mercier, Editor, W3C Marketing & Communications $Id: Overview.html,v 1.60 2019/10/15 12:05:52 coralie Exp $ Copyright © 2019 W3C ® (MIT, ERCIM, Keio, Beihang) Usage policies apply.
FRENKIE-CHIANG / DanmakuTPPBench[NeurIPS 2025] DanmakuTPPBench: A Multi-modal Benchmark for Temporal Point Process Modeling and Understanding
lindermanlab / PPSeq.jlNeyman-Scott point process model to identify sequential firing patterns in high-dimensional spike trains
omitakahiro / NeuralNetworkPointProcesscode for "Fully Neural Network based Model for General Temporal Point Processes"
xiaoshuai09 / Recurrent Point ProcessModeling the asynchronous event sequence via Recurrent Point Process
akharroubi / 3D Data Essentials3D DATA PROCESSING AND MODELING ESSENTIALS: A curated collection of tools, literature, courses, and datasets for mastering 3D point cloud technology.
hpaulkeeler / PostsPieces of code that have appeared on my blog.
givgramacho / CERN Quantum Computing CourseQuantum computing is one the most promising new trends in information processing. In this course, we will introduce from scratch the basic concepts of the quantum circuit model (qubits, gates and measures) and use them to study some of the most important quantum algorithms and protocols, including those that can be implemented with a few qubits (BB84, quantum teleportation, superdense coding...) as well as those that require multi-qubit systems (Deutsch-Jozsa, Grover, Shor..). We will also cover some of the most recent applications of quantum computing in the fields of optimization and simulation (with special emphasis on the use of quantum annealing, the quantum approximate optimization algorithm and the variational quantum eigensolver) and quantum machine learning (for instance, through the use of quantum support vector machines and quantum variational classifiers). We will also give examples of how these techniques can be used in chemistry simulations and high energy physics problems. The focus of the course will be on the practical aspects of quantum computing and on the implementation of algorithms in quantum simulators and actual quantum computers (as the ones available on the IBM Quantum Experience and D-Wave Leap). No previous knowledge of quantum physics is required and, from the mathematical point of view, only a good command of basic linear algebra is assumed. Some familiarity with the python programming language would be helpful, but is not required either.
hisrg / SNPESnapdragon Neural Processing Engine (SNPE) SDKThe Snapdragon Neural Processing Engine (SNPE) is a Qualcomm Snapdragon software accelerated runtime for the execution of deep neural networks. With SNPE, users can: Execute an arbitrarily deep neural network Execute the network on the SnapdragonTM CPU, the AdrenoTM GPU or the HexagonTM DSP. Debug the network execution on x86 Ubuntu Linux Convert Caffe, Caffe2, ONNXTM and TensorFlowTM models to a SNPE Deep Learning Container (DLC) file Quantize DLC files to 8 bit fixed point for running on the Hexagon DSP Debug and analyze the performance of the network with SNPE tools Integrate a network into applications and other code via C++ or Java
Aryia-Behroziuan / Other SourcesAsada, M.; Hosoda, K.; Kuniyoshi, Y.; Ishiguro, H.; Inui, T.; Yoshikawa, Y.; Ogino, M.; Yoshida, C. (2009). "Cognitive developmental robotics: a survey". IEEE Transactions on Autonomous Mental Development. 1 (1): 12–34. doi:10.1109/tamd.2009.2021702. S2CID 10168773. "ACM Computing Classification System: Artificial intelligence". ACM. 1998. Archived from the original on 12 October 2007. Retrieved 30 August 2007. Goodman, Joanna (2016). Robots in Law: How Artificial Intelligence is Transforming Legal Services (1st ed.). Ark Group. ISBN 978-1-78358-264-8. Archived from the original on 8 November 2016. Retrieved 7 November 2016. Albus, J. S. (2002). "4-D/RCS: A Reference Model Architecture for Intelligent Unmanned Ground Vehicles" (PDF). In Gerhart, G.; Gunderson, R.; Shoemaker, C. (eds.). Proceedings of the SPIE AeroSense Session on Unmanned Ground Vehicle Technology. Unmanned Ground Vehicle Technology IV. 3693. pp. 11–20. Bibcode:2002SPIE.4715..303A. CiteSeerX 10.1.1.15.14. doi:10.1117/12.474462. S2CID 63339739. Archived from the original (PDF) on 25 July 2004. Aleksander, Igor (1995). Artificial Neuroconsciousness: An Update. IWANN. Archived from the original on 2 March 1997. BibTex Archived 2 March 1997 at the Wayback Machine. Bach, Joscha (2008). "Seven Principles of Synthetic Intelligence". In Wang, Pei; Goertzel, Ben; Franklin, Stan (eds.). Artificial General Intelligence, 2008: Proceedings of the First AGI Conference. IOS Press. pp. 63–74. ISBN 978-1-58603-833-5. Archived from the original on 8 July 2016. Retrieved 16 February 2016. "Robots could demand legal rights". BBC News. 21 December 2006. Archived from the original on 15 October 2019. Retrieved 3 February 2011. Brooks, Rodney (1990). "Elephants Don't Play Chess" (PDF). Robotics and Autonomous Systems. 6 (1–2): 3–15. CiteSeerX 10.1.1.588.7539. doi:10.1016/S0921-8890(05)80025-9. Archived (PDF) from the original on 9 August 2007. Brooks, R. A. (1991). "How to build complete creatures rather than isolated cognitive simulators". In VanLehn, K. (ed.). Architectures for Intelligence. Hillsdale, NJ: Lawrence Erlbaum Associates. pp. 225–239. CiteSeerX 10.1.1.52.9510. Buchanan, Bruce G. (2005). "A (Very) Brief History of Artificial Intelligence" (PDF). AI Magazine: 53–60. Archived from the original (PDF) on 26 September 2007. Butler, Samuel (13 June 1863). "Darwin among the Machines". Letters to the Editor. The Press. Christchurch, New Zealand. Archived from the original on 19 September 2008. Retrieved 16 October 2014 – via Victoria University of Wellington. Clark, Jack (8 December 2015). "Why 2015 Was a Breakthrough Year in Artificial Intelligence". Bloomberg News. Archived from the original on 23 November 2016. Retrieved 23 November 2016. After a half-decade of quiet breakthroughs in artificial intelligence, 2015 has been a landmark year. Computers are smarter and learning faster than ever. "AI set to exceed human brain power". CNN. 26 July 2006. Archived from the original on 19 February 2008. Dennett, Daniel (1991). Consciousness Explained. The Penguin Press. ISBN 978-0-7139-9037-9. Domingos, Pedro (2015). The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World. Basic Books. ISBN 978-0-465-06192-1. Dowe, D. L.; Hajek, A. R. (1997). "A computational extension to the Turing Test". Proceedings of the 4th Conference of the Australasian Cognitive Science Society. Archived from the original on 28 June 2011. Dreyfus, Hubert (1972). What Computers Can't Do. New York: MIT Press. ISBN 978-0-06-011082-6. Dreyfus, Hubert; Dreyfus, Stuart (1986). Mind over Machine: The Power of Human Intuition and Expertise in the Era of the Computer. Oxford, UK: Blackwell. ISBN 978-0-02-908060-3. Archived from the original on 26 July 2020. Retrieved 22 August 2020. Dreyfus, Hubert (1992). What Computers Still Can't Do. New York: MIT Press. ISBN 978-0-262-54067-4. Dyson, George (1998). Darwin among the Machines. Allan Lane Science. ISBN 978-0-7382-0030-9. Archived from the original on 26 July 2020. Retrieved 22 August 2020. Edelman, Gerald (23 November 2007). "Gerald Edelman – Neural Darwinism and Brain-based Devices". Talking Robots. Archived from the original on 8 October 2009. Edelson, Edward (1991). The Nervous System. New York: Chelsea House. ISBN 978-0-7910-0464-7. Archived from the original on 26 July 2020. Retrieved 18 November 2019. Fearn, Nicholas (2007). The Latest Answers to the Oldest Questions: A Philosophical Adventure with the World's Greatest Thinkers. New York: Grove Press. ISBN 978-0-8021-1839-4. Gladwell, Malcolm (2005). Blink. New York: Little, Brown and Co. ISBN 978-0-316-17232-5. Gödel, Kurt (1951). Some basic theorems on the foundations of mathematics and their implications. Gibbs Lecture. In Feferman, Solomon, ed. (1995). Kurt Gödel: Collected Works, Vol. III: Unpublished Essays and Lectures. Oxford University Press. pp. 304–23. ISBN 978-0-19-514722-3. Haugeland, John (1985). Artificial Intelligence: The Very Idea. Cambridge, Mass.: MIT Press. ISBN 978-0-262-08153-5. Hawkins, Jeff; Blakeslee, Sandra (2005). On Intelligence. New York, NY: Owl Books. ISBN 978-0-8050-7853-4. Henderson, Mark (24 April 2007). "Human rights for robots? We're getting carried away". The Times Online. London. Archived from the original on 31 May 2014. Retrieved 31 May 2014. Hernandez-Orallo, Jose (2000). "Beyond the Turing Test". Journal of Logic, Language and Information. 9 (4): 447–466. doi:10.1023/A:1008367325700. S2CID 14481982. Hernandez-Orallo, J.; Dowe, D. L. (2010). "Measuring Universal Intelligence: Towards an Anytime Intelligence Test". Artificial Intelligence. 174 (18): 1508–1539. CiteSeerX 10.1.1.295.9079. doi:10.1016/j.artint.2010.09.006. Hinton, G. E. (2007). "Learning multiple layers of representation". Trends in Cognitive Sciences. 11 (10): 428–434. doi:10.1016/j.tics.2007.09.004. PMID 17921042. S2CID 15066318. Hofstadter, Douglas (1979). Gödel, Escher, Bach: an Eternal Golden Braid. New York, NY: Vintage Books. ISBN 978-0-394-74502-2. Holland, John H. (1975). Adaptation in Natural and Artificial Systems. University of Michigan Press. ISBN 978-0-262-58111-0. Archived from the original on 26 July 2020. Retrieved 17 December 2019. Howe, J. (November 1994). "Artificial Intelligence at Edinburgh University: a Perspective". Archived from the original on 15 May 2007. Retrieved 30 August 2007. Hutter, M. (2012). "One Decade of Universal Artificial Intelligence". Theoretical Foundations of Artificial General Intelligence. Atlantis Thinking Machines. 4. pp. 67–88. CiteSeerX 10.1.1.228.8725. doi:10.2991/978-94-91216-62-6_5. ISBN 978-94-91216-61-9. S2CID 8888091. Kahneman, Daniel; Slovic, D.; Tversky, Amos (1982). Judgment under uncertainty: Heuristics and biases. Science. 185. New York: Cambridge University Press. pp. 1124–31. doi:10.1126/science.185.4157.1124. ISBN 978-0-521-28414-1. PMID 17835457. S2CID 143452957. Kaplan, Andreas; Haenlein, Michael (2019). "Siri, Siri in my Hand, who's the Fairest in the Land? On the Interpretations, Illustrations and Implications of Artificial Intelligence". Business Horizons. 62: 15–25. doi:10.1016/j.bushor.2018.08.004. Katz, Yarden (1 November 2012). "Noam Chomsky on Where Artificial Intelligence Went Wrong". The Atlantic. Archived from the original on 28 February 2019. Retrieved 26 October 2014. "Kismet". MIT Artificial Intelligence Laboratory, Humanoid Robotics Group. Archived from the original on 17 October 2014. Retrieved 25 October 2014. Koza, John R. (1992). Genetic Programming (On the Programming of Computers by Means of Natural Selection). MIT Press. Bibcode:1992gppc.book.....K. ISBN 978-0-262-11170-6. Kolata, G. (1982). "How can computers get common sense?". Science. 217 (4566): 1237–1238. Bibcode:1982Sci...217.1237K. doi:10.1126/science.217.4566.1237. PMID 17837639. Kumar, Gulshan; Kumar, Krishan (2012). "The Use of Artificial-Intelligence-Based Ensembles for Intrusion Detection: A Review". Applied Computational Intelligence and Soft Computing. 2012: 1–20. doi:10.1155/2012/850160. Kurzweil, Ray (1999). The Age of Spiritual Machines. Penguin Books. ISBN 978-0-670-88217-5. Kurzweil, Ray (2005). The Singularity is Near. Penguin Books. ISBN 978-0-670-03384-3. Lakoff, George; Núñez, Rafael E. (2000). Where Mathematics Comes From: How the Embodied Mind Brings Mathematics into Being. Basic Books. ISBN 978-0-465-03771-1. Langley, Pat (2011). "The changing science of machine learning". Machine Learning. 82 (3): 275–279. doi:10.1007/s10994-011-5242-y. Law, Diane (June 1994). Searle, Subsymbolic Functionalism and Synthetic Intelligence (Technical report). University of Texas at Austin. p. AI94-222. CiteSeerX 10.1.1.38.8384. Legg, Shane; Hutter, Marcus (15 June 2007). A Collection of Definitions of Intelligence (Technical report). IDSIA. arXiv:0706.3639. Bibcode:2007arXiv0706.3639L. 07-07. Lenat, Douglas; Guha, R. V. (1989). Building Large Knowledge-Based Systems. Addison-Wesley. ISBN 978-0-201-51752-1. Lighthill, James (1973). "Artificial Intelligence: A General Survey". Artificial Intelligence: a paper symposium. Science Research Council. Lucas, John (1961). "Minds, Machines and Gödel". In Anderson, A.R. (ed.). Minds and Machines. Archived from the original on 19 August 2007. Retrieved 30 August 2007. Lungarella, M.; Metta, G.; Pfeifer, R.; Sandini, G. (2003). "Developmental robotics: a survey". Connection Science. 15 (4): 151–190. CiteSeerX 10.1.1.83.7615. doi:10.1080/09540090310001655110. S2CID 1452734. Maker, Meg Houston (2006). "AI@50: AI Past, Present, Future". Dartmouth College. Archived from the original on 3 January 2007. Retrieved 16 October 2008. Markoff, John (16 February 2011). "Computer Wins on 'Jeopardy!': Trivial, It's Not". The New York Times. Archived from the original on 22 October 2014. Retrieved 25 October 2014. McCarthy, John; Minsky, Marvin; Rochester, Nathan; Shannon, Claude (1955). "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence". Archived from the original on 26 August 2007. Retrieved 30 August 2007.. McCarthy, John; Hayes, P. J. (1969). "Some philosophical problems from the standpoint of artificial intelligence". Machine Intelligence. 4: 463–502. CiteSeerX 10.1.1.85.5082. Archived from the original on 10 August 2007. Retrieved 30 August 2007. McCarthy, John (12 November 2007). "What Is Artificial Intelligence?". Archived from the original on 18 November 2015. Minsky, Marvin (1967). Computation: Finite and Infinite Machines. Englewood Cliffs, N.J.: Prentice-Hall. ISBN 978-0-13-165449-5. Archived from the original on 26 July 2020. Retrieved 18 November 2019. Minsky, Marvin (2006). The Emotion Machine. New York, NY: Simon & Schusterl. ISBN 978-0-7432-7663-4. Moravec, Hans (1988). Mind Children. Harvard University Press. ISBN 978-0-674-57616-2. Archived from the original on 26 July 2020. Retrieved 18 November 2019. Norvig, Peter (25 June 2012). "On Chomsky and the Two Cultures of Statistical Learning". Peter Norvig. Archived from the original on 19 October 2014. NRC (United States National Research Council) (1999). "Developments in Artificial Intelligence". Funding a Revolution: Government Support for Computing Research. National Academy Press. Needham, Joseph (1986). Science and Civilization in China: Volume 2. Caves Books Ltd. Newell, Allen; Simon, H. A. (1976). "Computer Science as Empirical Inquiry: Symbols and Search". Communications of the ACM. 19 (3): 113–126. doi:10.1145/360018.360022.. Nilsson, Nils (1983). "Artificial Intelligence Prepares for 2001" (PDF). AI Magazine. 1 (1). Archived (PDF) from the original on 17 August 2020. Retrieved 22 August 2020. Presidential Address to the Association for the Advancement of Artificial Intelligence. O'Brien, James; Marakas, George (2011). Management Information Systems (10th ed.). McGraw-Hill/Irwin. ISBN 978-0-07-337681-3. O'Connor, Kathleen Malone (1994). "The alchemical creation of life (takwin) and other concepts of Genesis in medieval Islam". University of Pennsylvania: 1–435. Archived from the original on 5 December 2019. Retrieved 27 August 2008. Oudeyer, P-Y. (2010). "On the impact of robotics in behavioral and cognitive sciences: from insect navigation to human cognitive development" (PDF). IEEE Transactions on Autonomous Mental Development. 2 (1): 2–16. doi:10.1109/tamd.2009.2039057. S2CID 6362217. Archived (PDF) from the original on 3 October 2018. Retrieved 4 June 2013. Penrose, Roger (1989). The Emperor's New Mind: Concerning Computer, Minds and The Laws of Physics. Oxford University Press. ISBN 978-0-19-851973-7. Poli, R.; Langdon, W. B.; McPhee, N. F. (2008). A Field Guide to Genetic Programming. Lulu.com. ISBN 978-1-4092-0073-4. Archived from the original on 8 August 2015. Retrieved 21 April 2008 – via gp-field-guide.org.uk. Rajani, Sandeep (2011). "Artificial Intelligence – Man or Machine" (PDF). International Journal of Information Technology and Knowledge Management. 4 (1): 173–176. Archived from the original (PDF) on 18 January 2013. Ronald, E. M. A. and Sipper, M. Intelligence is not enough: On the socialization of talking machines, Minds and Machines Archived 25 July 2020 at the Wayback Machine, vol. 11, no. 4, pp. 567–576, November 2001. Ronald, E. M. A. and Sipper, M. What use is a Turing chatterbox? Archived 25 July 2020 at the Wayback Machine, Communications of the ACM, vol. 43, no. 10, pp. 21–23, October 2000. "Science". August 1982. Archived from the original on 25 July 2020. Retrieved 16 February 2016. Searle, John (1980). "Minds, Brains and Programs" (PDF). Behavioral and Brain Sciences. 3 (3): 417–457. doi:10.1017/S0140525X00005756. Archived (PDF) from the original on 17 March 2019. Retrieved 22 August 2020. Searle, John (1999). Mind, language and society. New York, NY: Basic Books. ISBN 978-0-465-04521-1. OCLC 231867665. Archived from the original on 26 July 2020. Retrieved 22 August 2020. Shapiro, Stuart C. (1992). "Artificial Intelligence". In Shapiro, Stuart C. (ed.). Encyclopedia of Artificial Intelligence (PDF) (2nd ed.). New York: John Wiley. pp. 54–57. ISBN 978-0-471-50306-4. Archived (PDF) from the original on 1 February 2016. Retrieved 29 May 2009. Simon, H. A. (1965). The Shape of Automation for Men and Management. New York: Harper & Row. Archived from the original on 26 July 2020. Retrieved 18 November 2019. Skillings, Jonathan (3 July 2006). "Getting Machines to Think Like Us". cnet. Archived from the original on 16 November 2011. Retrieved 3 February 2011. Solomonoff, Ray (1956). An Inductive Inference Machine (PDF). Dartmouth Summer Research Conference on Artificial Intelligence. Archived (PDF) from the original on 26 April 2011. Retrieved 22 March 2011 – via std.com, pdf scanned copy of the original. Later published as Solomonoff, Ray (1957). "An Inductive Inference Machine". IRE Convention Record. Section on Information Theory, part 2. pp. 56–62. Tao, Jianhua; Tan, Tieniu (2005). Affective Computing and Intelligent Interaction. Affective Computing: A Review. LNCS 3784. Springer. pp. 981–995. doi:10.1007/11573548. Tecuci, Gheorghe (March–April 2012). "Artificial Intelligence". Wiley Interdisciplinary Reviews: Computational Statistics. 4 (2): 168–180. doi:10.1002/wics.200. Thro, Ellen (1993). Robotics: The Marriage of Computers and Machines. New York: Facts on File. ISBN 978-0-8160-2628-9. Archived from the original on 26 July 2020. Retrieved 22 August 2020. Turing, Alan (October 1950), "Computing Machinery and Intelligence", Mind, LIX (236): 433–460, doi:10.1093/mind/LIX.236.433, ISSN 0026-4423. van der Walt, Christiaan; Bernard, Etienne (2006). "Data characteristics that determine classifier performance" (PDF). Archived from the original (PDF) on 25 March 2009. Retrieved 5 August 2009. Vinge, Vernor (1993). "The Coming Technological Singularity: How to Survive in the Post-Human Era". Vision 21: Interdisciplinary Science and Engineering in the Era of Cyberspace: 11. Bibcode:1993vise.nasa...11V. Archived from the original on 1 January 2007. Retrieved 14 November 2011. Wason, P. C.; Shapiro, D. (1966). "Reasoning". In Foss, B. M. (ed.). New horizons in psychology. Harmondsworth: Penguin. Archived from the original on 26 July 2020. Retrieved 18 November 2019. Weizenbaum, Joseph (1976). Computer Power and Human Reason. San Francisco: W.H. Freeman & Company. ISBN 978-0-7167-0464-5. Weng, J.; McClelland; Pentland, A.; Sporns, O.; Stockman, I.; Sur, M.; Thelen, E. (2001). "Autonomous mental development by robots and animals" (PDF). Science. 291 (5504): 599–600. doi:10.1126/science.291.5504.599. PMID 11229402. S2CID 54131797. Archived (PDF) from the original on 4 September 2013. Retrieved 4 June 2013 – via msu.edu. "Applications of AI". www-formal.stanford.edu. Archived from the original on 28 August 2016. Retrieved 25 September 2016. Further reading DH Author, 'Why Are There Still So Many Jobs? The History and Future of Workplace Automation' (2015) 29(3) Journal of Economic Perspectives 3. Boden, Margaret, Mind As Machine, Oxford University Press, 2006. Cukier, Kenneth, "Ready for Robots? How to Think about the Future of AI", Foreign Affairs, vol. 98, no. 4 (July/August 2019), pp. 192–98. George Dyson, historian of computing, writes (in what might be called "Dyson's Law") that "Any system simple enough to be understandable will not be complicated enough to behave intelligently, while any system complicated enough to behave intelligently will be too complicated to understand." (p. 197.) Computer scientist Alex Pentland writes: "Current AI machine-learning algorithms are, at their core, dead simple stupid. They work, but they work by brute force." (p. 198.) Domingos, Pedro, "Our Digital Doubles: AI will serve our species, not control it", Scientific American, vol. 319, no. 3 (September 2018), pp. 88–93. Gopnik, Alison, "Making AI More Human: Artificial intelligence has staged a revival by starting to incorporate what we know about how children learn", Scientific American, vol. 316, no. 6 (June 2017), pp. 60–65. Johnston, John (2008) The Allure of Machinic Life: Cybernetics, Artificial Life, and the New AI, MIT Press. Koch, Christof, "Proust among the Machines", Scientific American, vol. 321, no. 6 (December 2019), pp. 46–49. Christof Koch doubts the possibility of "intelligent" machines attaining consciousness, because "[e]ven the most sophisticated brain simulations are unlikely to produce conscious feelings." (p. 48.) According to Koch, "Whether machines can become sentient [is important] for ethical reasons. If computers experience life through their own senses, they cease to be purely a means to an end determined by their usefulness to... humans. Per GNW [the Global Neuronal Workspace theory], they turn from mere objects into subjects... with a point of view.... Once computers' cognitive abilities rival those of humanity, their impulse to push for legal and political rights will become irresistible – the right not to be deleted, not to have their memories wiped clean, not to suffer pain and degradation. The alternative, embodied by IIT [Integrated Information Theory], is that computers will remain only supersophisticated machinery, ghostlike empty shells, devoid of what we value most: the feeling of life itself." (p. 49.) Marcus, Gary, "Am I Human?: Researchers need new ways to distinguish artificial intelligence from the natural kind", Scientific American, vol. 316, no. 3 (March 2017), pp. 58–63. A stumbling block to AI has been an incapacity for reliable disambiguation. An example is the "pronoun disambiguation problem": a machine has no way of determining to whom or what a pronoun in a sentence refers. (p. 61.) E McGaughey, 'Will Robots Automate Your Job Away? Full Employment, Basic Income, and Economic Democracy' (2018) SSRN, part 2(3) Archived 24 May 2018 at the Wayback Machine. George Musser, "Artificial Imagination: How machines could learn creativity and common sense, among other human qualities", Scientific American, vol. 320, no. 5 (May 2019), pp. 58–63. Myers, Courtney Boyd ed. (2009). "The AI Report" Archived 29 July 2017 at the Wayback Machine. Forbes June 2009 Raphael, Bertram (1976). The Thinking Computer. W.H.Freeman and Company. ISBN 978-0-7167-0723-3. Archived from the original on 26 July 2020. Retrieved 22 August 2020. Scharre, Paul, "Killer Apps: The Real Dangers of an AI Arms Race", Foreign Affairs, vol. 98, no. 3 (May/June 2019), pp. 135–44. "Today's AI technologies are powerful but unreliable. Rules-based systems cannot deal with circumstances their programmers did not anticipate. Learning systems are limited by the data on which they were trained. AI failures have already led to tragedy. Advanced autopilot features in cars, although they perform well in some circumstances, have driven cars without warning into trucks, concrete barriers, and parked cars. In the wrong situation, AI systems go from supersmart to superdumb in an instant. When an enemy is trying to manipulate and hack an AI system, the risks are even greater." (p. 140.) Serenko, Alexander (2010). "The development of an AI journal ranking based on the revealed preference approach" (PDF). Journal of Informetrics. 4 (4): 447–459. doi:10.1016/j.joi.2010.04.001. Archived (PDF) from the original on 4 October 2013. Retrieved 24 August 2013. Serenko, Alexander; Michael Dohan (2011). "Comparing the expert survey and citation impact journal ranking methods: Example from the field of Artificial Intelligence" (PDF). Journal of Informetrics. 5 (4): 629–649. doi:10.1016/j.joi.2011.06.002. Archived (PDF) from the original on 4 October 2013. Retrieved 12 September 2013. Sun, R. & Bookman, L. (eds.), Computational Architectures: Integrating Neural and Symbolic Processes. Kluwer Academic Publishers, Needham, MA. 1994. Tom Simonite (29 December 2014). "2014 in Computing: Breakthroughs in Artificial Intelligence". MIT Technology Review. Tooze, Adam, "Democracy and Its Discontents", The New York Review of Books, vol. LXVI, no. 10 (6 June 2019), pp. 52–53, 56–57. "Democracy has no clear answer for the mindless operation of bureaucratic and technological power. We may indeed be witnessing its extension in the form of artificial intelligence and robotics. Likewise, after decades of dire warning, the environmental problem remains fundamentally unaddressed.... Bureaucratic overreach and environmental catastrophe are precisely the kinds of slow-moving existential challenges that democracies deal with very badly.... Finally, there is the threat du jour: corporations and the technologies they promote." (pp. 56–57.)
sduprey / Optimal Transaction ExecutionThis entry contains two topics The first item is entirely based on the following paper: http://sfb649.wiwi.hu-berlin.de/papers/pdf/SFB649DP2011-056.pdf It contains 2 MATLAB demonstrating script : DATA_preprocessing.m & VAR_modeling_script.m DATA_preprocessing.m uses the LOBSTER framework (https://lobster.wiwi.hu-berlin.de/) to preprocess high frequency data from the NASDAQ Total View ITCH (csv files) allowing us to reconstruct exactly at each time the order book up to ten depths. Just look at the published script ! VAR_modeling_script.m contains the modeling of the whole order book as VEC/VAR process. It uses the great VAR/VEC Joahnsen cointegration framework. After calibrating your VAR model, you then assess the impact of an order using shock scenario (sensitivity analysis) to the VAR process. We deal with 3 scenarii : normal limit order, aggressive limit order & normal market order). Play section by section the script (to open up figures which contain a lot of graphs). It contains a power point to help you present this complex topic. The second item is entirely based on the following paper : http://www.courant.nyu.edu/~almgren/papers/optliq.pdf It contains a mupad document : symbolic_demo.mn I did struggle to get something nice with the symbolic toolbox. I was not able to drive a continuous workflow and had to recode some equations myself. I nevertheless managed to get a closed form solution for the simplified linear cost model. It contains a MATLAB demonstrating script : working_script.m For more sophisticated cost model, there is no more closed form and we there highlighted MATLAB numerical optimization abilities (fmincon). It contains an Optimization Apps you can install. Just launch the optimization with the default parameters. And then switch the slider between volatility risk and liquidation costs to see the trading strategies evolve on the efficient frontier. It contains a power point to help you present this complex topic.
Zhiyuan-R / Tiger DiffusionThis is the official code for the CVPR 2024 Publication: Tiger: Time-Varying Denoising Model for 3D Point Cloud Generation with Diffusion Process
mtth-bfft / NtsecStandalone tool to explore the security model of Windows and its NT kernel. Use it to introspect privilege assignments and access right assignments, enumerate attack surfaces from the point of view of a sandboxed process, etc.
QibangLiu / GINOTGINOT is a deep learning model that combines transformers with neural operators for accurate forward predictions on arbitrary 2D and 3D geometries. It processes surface point clouds using attention-based encoding with sampling and grouping, ensuring robustness to point density, order invariance, and padding resilience.
SOYJUN / Application With Raw IP SocketsOverview For this assignment you will be developing an application that uses raw IP sockets to ‘walk’ around an ordered list of nodes (given as a command line argument at the ‘source’ node, which is the node at which the tour was initiated), in a manner similar to the IP SSRR (Strict Source and Record Route) option. At each node, the application pings the preceding node in the tour. However, unlike the ping code in Stevens, you will be sending the ping ICMP echo request messages through a SOCK_RAW-type PF_PACKET socket and implementing ARP functionality to find the Ethernet address of the target node. Finally, when the ‘walk’ is completed, the group of nodes visited on the tour will exchange multicast messages. Your code will consist of two process modules, a ‘Tour’ application module (which will implement all the functionality outlined above, except for ARP activity) and an ARP module. The following should prove to be useful reference material for the assignment: Sections 21.2, 21.3, 21.6 and 21.10, Chapter 21, on Multicasting. Sections 27.1 to 27.3, Chapter 27, on the IP SSRR option. Sections 28.1 to 28.5, Chapter 28, on raw sockets, the IP_HDRINCL socket option, and ping. Sections 15.5, Chapter 15, on Unix domain SOCK_STREAM sockets. Figure 29.14, p. 807, and the corresponding explanation on p. 806, on filling in an IP header when the IP_HDRINCL socket option is in effect. The Lecture Slides on ARP & RARP (especially Section 4.4, ARP Packet Format, and the Figure 4.3 it includes). The link http://www.pdbuchan.com/rawsock/rawsock.html contains useful code samples that use IP raw sockets and PF_PACKET sockets. Note, in partcular, the code “icmp4_ll.c” in Table 2 for building an echo request sent through a PF_PACKET SOCK_RAW socket. The VMware environment You will be using the same vm1 , . . . . . , vm10 nodes you used for Assignment 3. However, unlike Assignment 3, you should use only interfaces eth0 and their associated IP addresses and ignore the other Ethernet interfaces that nodes have (interfaces eth0 make vm1 , . . . . . , vm10 look as if they belong to the same Ethernet LAN segment IP network 130.245.156.0/24). Note that, apart from the primary IP addresses associated with interfaces eth0, some nodes might also have one or more alias IP addresses associated with their interface eth0. Tour application module specifications The application will create a total of four sockets: two IP raw sockets, a PF_PACKET socket and a UDP socket for multicasting. We shall call the two IP raw sockets the ‘rt ’ (‘route traversal’) and ‘pg ’ (‘ping’) sockets, respectively. The rt socket should have the IP_HDRINCL option set. You will only be receiving ICMP echo reply messages through the pg socket (and not sending echo requests), so it does not matter whether it has the IP_HDRINCL option set or not. The pg socket should have protocol value (i.e., protocol demultiplexing key in the IP header) IPPROTO_ICMP. The rt socket should have a protocol value that identifies the application - i.e., some value other than the IPPROTO_XXXX values in /usr/include/netinet/in.h. However, remember that you will all be running your code using the same root account on the vm1 , . . . . . , vm10 nodes. So if two of you happen to choose the same protocol value and happen to be running on the same vm node at the same time, your applications will receive each other’s IP packets. For that reason, try to choose a protocol value for your rt socket that is likely to be unique to yourself. The PF_PACKET socket should be of type SOCK_RAW (not SOCK_DGRAM). This socket should have a protocol value of ETH_P_IP = 0x0800 (IPv4). The UDP socket for multicasting will be discussed below. Note that, depending on how you choose to bind that socket, you might actually need to have two UDP sockets for multicast communication – see bottom of p. 576, Section 21.10. Your application will, of course, have to be running on every vm node that is included in the tour. When evoking the application on the source node, the user supplies a sequence of vm node names (not IP addresses) to be visited in order. This command line sequence starts with the next node to be visited from the source node (i.e., it does not start with the source node itself). The sequence can include any number of repeated visits to the same node. For example, suppose that the source node is vm3 and the executable is called badr_tour : [root@vm3/root]# badr_tour vm2 vm10 vm4 vm7 vm5 vm2 vm6 vm2 vm9 vm4 vm7 vm2 vm6 vm5 vm1 vm10 vm8 (but note that the tour does not necessarily have to visit every vm node; and the same node should not appear consequentively in the tour list – i.e., the next node on the tour cannot be the current node itself). The application turns the sequence into a list of IP addresses for source routing. It also adds the IP address of the source node itself to the beginning of the list. The list thus produced will be carried as the payload of an IP packet, not as a SSRR option in the packet header. It is our application which will ensure that every node in the sequence is visited in order, not the IP SSRR capability. The source node should also add to the list an IP multicast address and a port number of its choice. It should also join the multicast group at that address and port number on its UDP socket. The TTL for outgoing multicasts should be set to 1. The application then fills in the header of an IP packet, designating itself as the IP source, and the next node to be visited as the IP destination. The packet is sent out on the rt socket. Note that on Linux, all the fields of the packet header must be in network byte order (Stevens, Section 28.3, p. 737, the fourth bullet point). When filling in the packet header, you should explicitly fill in the identification field (recall that, with the IP_HDRINCL socket option, if the identification field is given value 0, then the kernel will set its value). Try to make sure that the value you choose is likely to be unique to yourself (for reasons similar to those explained with respect to the IPPROTO_XXXX in 1. above). When a node receives an IP packet on its rt socket, it should first check that the identification field carries the right value (this implies that you will hard code your choice of identification field value determined in item 2 above in your code). If the identification field value does not check out, the packet is ignored. For a valid packet : Print out a message along the lines of: <time> received source routing packet from <hostname> <time> is the current time in human-readable format (see lines 19 & 20 in Figure 1.9, p. 14, and the corresponding explanation on p. 14f.), and <hostname> is the host name corresponding to the source IP address in the header of the received packet. If this is the first time the node is visited, the application should use the multicast address and port number in the packet received to join the multicast group on its UDP socket. The TTL for outgoing multicasts should be set to 1. The application updates the list in the payload, so that the next node in the tour can easily identify what the next hop from itself will be when it receives the packet. How you do this I leave up to you. You could, for example, include as part of the payload a pointer field into the list of nodes to be visited. This pointer would then be updated to the next entry in the list as the packet progresses hop by hop (see Figure 27.1 and the associated explanation on pp. 711-712). Other solutions are, of course, possible. The application then fills in a new IP header, designating itself as the IP source, and the next node to be visited as the IP destination. The identification field should be set to the same value as in the received packet. The packet is sent out on the rt socket. The node should also initiate pinging to the preceding node in the tour (the IP address of which it should pick up from the header of the received packet). However, unlike the Stevens ping code, it will be using the SOCK_RAW-type PF_PACKET socket of item 1 above to send the ICMP echo request messages. Before it can send echo request messages, the application has to call on the ARP module you will implement to get the Ethernet address of this preceding / ‘target’ node; this call is made using the API function areq which you will also implement (see sections ARP module specifications & API specifications below). Note that ARP has to be evoked every time the application wants to send out an echo request message, and not just the first time. An echo request message has to be encapsulated in a properly-formulated IP packet, which is in turn encapsulated in a properly-formulated Ethernet frame transmitted out through the PF_PACKET socket ; otherwise, ICMP at the source node will not receive it. You will have to modify Stevens’ ping code accordingly, specifically, the send_v4 function. In particular, the Ethernet frame must have a value of ETH_P_IP = 0x0800 (IPv4 – see <linux/if_ether.h>) in the frame type / ‘length’ field ; and the encapsulated IP packet must have a value of IPPROTO_ICMP = 0x01 (ICMPv4 – see <netinet_in.h>) in its protocol field. You should also simplify the ping code in its entirety by stripping all the ‘indirection’ IPv4 / IPv6 dual-operability paraphernalia and making the code work just for IPv4. Also note that the functions host_serv and freeaddrinfo, together with the associated structure addrinfo (see Sections 11.6, 11.8 & 11.11), in Figures 27.3, 27.6 & 28.5 ( pp. 713, 716 & 744f., respectively) can be replaced by the function gethostbyname and associated structure hostent (see Section 11.3) where needed. Also, there is no ‘-v’ verbose option, so this too should be stripped from Stevens’ code. When a node is ready to start pinging, it first prints out a ‘PING’ message similar to lines 32-33 of Figure 28.5, p. 744. It then builds up ICMP echo request messages and sends them to the source node every 1 second through the PF_PACKET socket. It also reads incoming echo response messages off the pg socket, in response to which it prints out the same kind of output as the code of Figure 28.8, p. 748. If this node and its preceding node have been previously visited in that order during the tour, then pinging would have already been initiated from the one to the other in response to the first visit, and nothing further should nor need be done during second and subsequent visits. In light of the above, note that once a node initiates pinging, it needs to read from both its rt and pg sockets, necessitating the use of the select function. As will be clear from what follows below, the application will anyway be needing also to simultaneously monitor its UDP socket for incoming multicast datagrams. When the last node on the tour is reached, and if this is the first time it is visited, it joins the multicast group and starts pinging the preceding node (if it is not already doing so). After a few echo replies are received (five, say), it sends out the multicast message below on its UDP socket (i.e., the node should wait about five seconds before sending the multicast message) : <<<<< This is node vmi . Tour has ended . Group members please identify yourselves. >>>>> where vmi is the name (not IP address) of the node. The node should also print this message out on stdout preceded, on the same line, by the phrase: Node vmi . Sending: <then print out the message sent>. Each node vmj receiving this message should print out the message received preceded, on the same line, by the phrase: Node vmj . Received <then print out the message received>. Each such node in step a above should then immediately stop its pinging activity. The node should then send out the following multicast message: <<<<< Node vmj . I am a member of the group. >>>>> and print out this message preceded, on the same line, by the phrase: Node vmj . Sending: <then print out the message sent>. Each node receiving these second multicast messages (i.e., the messages that nodes – including itself – sent out in step c above) should print each such message out preceded, on the same line, by the phrase: Node vmk . Received: <then print out the message received>. Reading from the socket in step d above should be implemented with a 5-second timeout. When the timeout expires, the node should print out another message to the effect that it is terminating the Tour application, and gracefully exit its Tour process. Note that under Multicast specifications, the last node in the tour, which sends out the End of Tour message, should itself receive a copy of that message and, when it does, it should behave exactly as do the other nodes in steps a. – e. above. ARP module specifications Your executable is evoked with no command line arguments. Like the Tour module, it will be running on every vm node. It uses the get_hw_addrs function of Assignment 3 to explore its node’s interfaces and build a set of <IP address , HW address> matching pairs for all eth0 interface IP addresses (including alias IP addresses, if any). Write out to stdout in some appropriately clear format the address pairs found. The module creates two sockets: a PF_PACKET socket and a Unix domain socket. The PF_PACKET should be of type SOCK_RAW (not type SOCK_DGRAM) with a protocol value of your choice (but not one of the standard values defined in <linux/if_ether.h>) which is, hopefully, unique to yourself. This value effectively becomes the protocol value for your implementation of ARP. Because this protocol value will be carried in the frame type / ‘length’ field of the Ethernet frame header (see Figure 4.3 of the ARP & RARP handout), the value chosen should be not less than 1536 (0x600) so that it is not misinterpreted as the length of an Ethernet 802.3 frame. The Unix domain socket should be of type SOCK_STREAM (not SOCK_DGRAM). It is a listening socket bound to a ‘well-known’ sun_path file. This socket will be used to communicate with the function areq that is implemented in the Tour module (see the section API specifications below). In this context, areq will act as the client and the ARP module as the server. The ARP module then sits in an infinite loop, monitoring these two sockets. As ARP request messages arrive on the PF_PACKET socket, the module processes them, and responds with ARP reply messages as appropriate. The protocol builds a ‘cache’ of matching <IP address , HW address> pairs from the replies (and requests – see below) it receives. For simplicity, and unlike the real ARP, we shall not implement timing out mechanisms for these cache entries. A cache entry has five parts: (i) IP address ; (ii) HW address ; (iii) sll_ifindex (the interface to be used for reaching the matching pair <(i) , (ii)>) ; (iv) sll_hatype ; and (v) a Unix-domain connection-socket descriptor for a connected client (see the section API specifications below for the latter three). When an ARP reply is being entered in the cache, the ARP module uses the socket descriptor in (v) to send a reply to the client, closes the connection socket, and deletes the socket descriptor from the cache entry. Note that, like the real ARP, when an ARP request is received by a node, and if the request pertains to that receiving node, the sender’s (see Figure 4.3 of the ARP & RARP handout) <IP address, HW address> matching pair should be entered into the cache if it is not already there (together, of course, with (iii) sll_ifindex & (iv) sll_hatype), or updated if need be if such an entry already exists in the cache. If the ARP request received does not pertain to the node receiving it, but there is already an entry in that receiving node's cache for the sender’s <IP address, HW address> matching pair, that entry should be checked and updated if need be. If there is no such entry, no action is taken (in particular, and unlike the case above, no new entry should be made in the receiving node's cache of the sender’s <IP address, HW address> matching pair if such an entry does not already exist). ARP request and reply messages have the same format as Figure 4.3 of the ARP & RARP handout, but with an extra 2-byte identification field added at the beginning which you fill with a value chosen so that it has a high probability of being unique to yourself. This value is to be echoed in the reply message, and helps to act as a further filter in case some other student happens to have fortuitously chosen the same value as yourself for the protocol parameter of the ARP PF_PACKET. Values in the fields of our ARP messages must be in network byte order. You might find the system header file <linux/if_arp.h> useful for manipulating ARP request and reply messages, but remember that our version of these messages have an extra two-byte field as mentioned above. Your code should print out on stdout, in some appropriately clear format, the contents of the Ethernet frame header and ARP request message you send. As described in Section 4.4 of the ARP & RARP handout, the node that responds to the request should, in its reply message, swap the two sender addresses with the two target addresses, as well as, of course, echo back the extra identification field sent with the request. The protocol at this responding node should print out, in an appropriately clear format, both the request frame (header and ARP message) it receives and the reply frame it sends. Similarly, the node that sent the request should print out the reply frame it receives. Finally, recall that the node issuing the request sends out a broadcast Ethernet frame, but the responding node replies with a unicast frame. API specifications The API is for communication between the Tour process and the ARP process. It consists of a single function, areq, implemented in the Tour module. areq is called by send_v4 function of the application every time the latter want to send out an ICMP echo request message: int areq (struct sockaddr *IPaddr, socklen_t sockaddrlen, struct hwaddr *HWaddr); IPaddr contains the primary or alias IPaddress of a ‘target’ node on the LAN for which the corresponding hardware address is being requested. hwaddr is a new structure (and not a pre-existing type) modeled on the sockaddr_ll of PF_PACKET; you will have to declare it in your code. It is used to return the requested hardware address to the caller of areq : structure hwaddr { int sll_ifindex; /* Interface number */ unsigned short sll_hatype; /* Hardware type */ unsigned char sll_halen; /* Length of address */ unsigned char sll_addr[8]; /* Physical layer address */ }; areq creates a Unix domain socket of type SOCK_STREAM and connects to the ‘well-known’ sun_path file of the ARP listening socket. It sends the IP address from parameter IPaddr and the information in the three fields of parameter HWaddr to ARP. It then blocks on a read awaiting a reply from ARP. This read should be backed up by a timeout since it is possible that no reply is received for the request. If a timeout occurs, areq should close the socket and return to its caller indicating failure (through its int return value). Your application code should print out on stdout, in some appropriately clear format, a notification every time areq is called, giving the IP address for which a HW address is being sought. It should similarly print out the result when the call to areq returns (HW address returned, or failure). When the ARP module receives a request for a HW address from areq through its Unix domain listening socket, it first checks if the required HW address is already in the cache. If so, it can respond immediately to the areq and close the Unix domain connection socket. Else : it makes an ‘incomplete’ entry in the cache, consisting of parts (i), (iii), (iv) and (v) ; puts out an ARP request message on the network on its PF_PACKET socket; and starts monitoring the areq connection socket for readability – if the areq client closes the connection socket (this would occur in response to a timeout in areq), ARP deletes the corresponding incomplete entry from the cache (and ignores any subsequent ARP reply from the network if such is received). On the other hand, if ARP receives a reply from the network, it updates the incomplete cache entry, responds to areq, and closes the connection socket.
