75 skills found · Page 1 of 3
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.
ManojKumarPatnaik / Major Project ListA list of practical projects that anyone can solve in any programming language (See solutions). These projects are divided into multiple categories, and each category has its own folder. To get started, simply fork this repo. CONTRIBUTING See ways of contributing to this repo. You can contribute solutions (will be published in this repo) to existing problems, add new projects, or remove existing ones. Make sure you follow all instructions properly. Solutions You can find implementations of these projects in many other languages by other users in this repo. Credits Problems are motivated by the ones shared at: Martyr2’s Mega Project List Rosetta Code Table of Contents Numbers Classic Algorithms Graph Data Structures Text Networking Classes Threading Web Files Databases Graphics and Multimedia Security Numbers Find PI to the Nth Digit - Enter a number and have the program generate PI up to that many decimal places. Keep a limit to how far the program will go. Find e to the Nth Digit - Just like the previous problem, but with e instead of PI. Enter a number and have the program generate e up to that many decimal places. Keep a limit to how far the program will go. Fibonacci Sequence - Enter a number and have the program generate the Fibonacci sequence to that number or to the Nth number. Prime Factorization - Have the user enter a number and find all Prime Factors (if there are any) and display them. Next Prime Number - Have the program find prime numbers until the user chooses to stop asking for the next one. Find Cost of Tile to Cover W x H Floor - Calculate the total cost of the tile it would take to cover a floor plan of width and height, using a cost entered by the user. Mortgage Calculator - Calculate the monthly payments of a fixed-term mortgage over given Nth terms at a given interest rate. Also, figure out how long it will take the user to pay back the loan. For added complexity, add an option for users to select the compounding interval (Monthly, Weekly, Daily, Continually). Change Return Program - The user enters a cost and then the amount of money given. The program will figure out the change and the number of quarters, dimes, nickels, pennies needed for the change. Binary to Decimal and Back Converter - Develop a converter to convert a decimal number to binary or a binary number to its decimal equivalent. Calculator - A simple calculator to do basic operators. Make it a scientific calculator for added complexity. Unit Converter (temp, currency, volume, mass, and more) - Converts various units between one another. The user enters the type of unit being entered, the type of unit they want to convert to, and then the value. The program will then make the conversion. Alarm Clock - A simple clock where it plays a sound after X number of minutes/seconds or at a particular time. Distance Between Two Cities - Calculates the distance between two cities and allows the user to specify a unit of distance. This program may require finding coordinates for the cities like latitude and longitude. Credit Card Validator - Takes in a credit card number from a common credit card vendor (Visa, MasterCard, American Express, Discoverer) and validates it to make sure that it is a valid number (look into how credit cards use a checksum). Tax Calculator - Asks the user to enter a cost and either a country or state tax. It then returns the tax plus the total cost with tax. Factorial Finder - The Factorial of a positive integer, n, is defined as the product of the sequence n, n-1, n-2, ...1, and the factorial of zero, 0, is defined as being 1. Solve this using both loops and recursion. Complex Number Algebra - Show addition, multiplication, negation, and inversion of complex numbers in separate functions. (Subtraction and division operations can be made with pairs of these operations.) Print the results for each operation tested. Happy Numbers - A happy number is defined by the following process. Starting with any positive integer, replace the number by the sum of the squares of its digits, and repeat the process until the number equals 1 (where it will stay), or it loops endlessly in a cycle which does not include 1. Those numbers for which this process ends in 1 are happy numbers, while those that do not end in 1 are unhappy numbers. Display an example of your output here. Find the first 8 happy numbers. Number Names - Show how to spell out a number in English. You can use a preexisting implementation or roll your own, but you should support inputs up to at least one million (or the maximum value of your language's default bounded integer type if that's less). Optional: Support for inputs other than positive integers (like zero, negative integers, and floating-point numbers). Coin Flip Simulation - Write some code that simulates flipping a single coin however many times the user decides. The code should record the outcomes and count the number of tails and heads. Limit Calculator - Ask the user to enter f(x) and the limit value, then return the value of the limit statement Optional: Make the calculator capable of supporting infinite limits. Fast Exponentiation - Ask the user to enter 2 integers a and b and output a^b (i.e. pow(a,b)) in O(LG n) time complexity. Classic Algorithms Collatz Conjecture - Start with a number n > 1. Find the number of steps it takes to reach one using the following process: If n is even, divide it by 2. If n is odd, multiply it by 3 and add 1. Sorting - Implement two types of sorting algorithms: Merge sort and bubble sort. Closest pair problem - The closest pair of points problem or closest pair problem is a problem of computational geometry: given n points in metric space, find a pair of points with the smallest distance between them. Sieve of Eratosthenes - The sieve of Eratosthenes is one of the most efficient ways to find all of the smaller primes (below 10 million or so). Graph Graph from links - Create a program that will create a graph or network from a series of links. Eulerian Path - Create a program that will take as an input a graph and output either an Eulerian path or an Eulerian cycle, or state that it is not possible. An Eulerian path starts at one node and traverses every edge of a graph through every node and finishes at another node. An Eulerian cycle is an eulerian Path that starts and finishes at the same node. Connected Graph - Create a program that takes a graph as an input and outputs whether every node is connected or not. Dijkstra’s Algorithm - Create a program that finds the shortest path through a graph using its edges. Minimum Spanning Tree - Create a program that takes a connected, undirected graph with weights and outputs the minimum spanning tree of the graph i.e., a subgraph that is a tree, contains all the vertices, and the sum of its weights is the least possible. Data Structures Inverted index - An Inverted Index is a data structure used to create full-text search. Given a set of text files, implement a program to create an inverted index. Also, create a user interface to do a search using that inverted index which returns a list of files that contain the query term/terms. The search index can be in memory. Text Fizz Buzz - Write a program that prints the numbers from 1 to 100. But for multiples of three print “Fizz” instead of the number and for the multiples of five print “Buzz”. For numbers which are multiples of both three and five print “FizzBuzz”. Reverse a String - Enter a string and the program will reverse it and print it out. Pig Latin - Pig Latin is a game of alterations played in the English language game. To create the Pig Latin form of an English word the initial consonant sound is transposed to the end of the word and an ay is affixed (Ex.: "banana" would yield anana-bay). Read Wikipedia for more information on rules. Count Vowels - Enter a string and the program counts the number of vowels in the text. For added complexity have it report a sum of each vowel found. Check if Palindrome - Checks if the string entered by the user is a palindrome. That is that it reads the same forwards as backward like “racecar” Count Words in a String - Counts the number of individual words in a string. For added complexity read these strings in from a text file and generate a summary. Text Editor - Notepad-style application that can open, edit, and save text documents. Optional: Add syntax highlighting and other features. RSS Feed Creator - Given a link to RSS/Atom Feed, get all posts and display them. Quote Tracker (market symbols etc) - A program that can go out and check the current value of stocks for a list of symbols entered by the user. The user can set how often the stocks are checked. For CLI, show whether the stock has moved up or down. Optional: If GUI, the program can show green up and red down arrows to show which direction the stock value has moved. Guestbook / Journal - A simple application that allows people to add comments or write journal entries. It can allow comments or not and timestamps for all entries. Could also be made into a shoutbox. Optional: Deploy it on Google App Engine or Heroku or any other PaaS (if possible, of course). Vigenere / Vernam / Ceasar Ciphers - Functions for encrypting and decrypting data messages. Then send them to a friend. Regex Query Tool - A tool that allows the user to enter a text string and then in a separate control enter a regex pattern. It will run the regular expression against the source text and return any matches or flag errors in the regular expression. Networking FTP Program - A file transfer program that can transfer files back and forth from a remote web sever. Bandwidth Monitor - A small utility program that tracks how much data you have uploaded and downloaded from the net during the course of your current online session. See if you can find out what periods of the day you use more and less and generate a report or graph that shows it. Port Scanner - Enter an IP address and a port range where the program will then attempt to find open ports on the given computer by connecting to each of them. On any successful connections mark the port as open. Mail Checker (POP3 / IMAP) - The user enters various account information include web server and IP, protocol type (POP3 or IMAP), and the application will check for email at a given interval. Country from IP Lookup - Enter an IP address and find the country that IP is registered in. Optional: Find the Ip automatically. Whois Search Tool - Enter an IP or host address and have it look it up through whois and return the results to you. Site Checker with Time Scheduling - An application that attempts to connect to a website or server every so many minute or a given time and check if it is up. If it is down, it will notify you by email or by posting a notice on the screen. Classes Product Inventory Project - Create an application that manages an inventory of products. Create a product class that has a price, id, and quantity on hand. Then create an inventory class that keeps track of various products and can sum up the inventory value. Airline / Hotel Reservation System - Create a reservation system that books airline seats or hotel rooms. It charges various rates for particular sections of the plane or hotel. For example, first class is going to cost more than a coach. Hotel rooms have penthouse suites which cost more. Keep track of when rooms will be available and can be scheduled. Company Manager - Create a hierarchy of classes - abstract class Employee and subclasses HourlyEmployee, SalariedEmployee, Manager, and Executive. Everyone's pay is calculated differently, research a bit about it. After you've established an employee hierarchy, create a Company class that allows you to manage the employees. You should be able to hire, fire, and raise employees. Bank Account Manager - Create a class called Account which will be an abstract class for three other classes called CheckingAccount, SavingsAccount, and BusinessAccount. Manage credits and debits from these accounts through an ATM-style program. Patient / Doctor Scheduler - Create a patient class and a doctor class. Have a doctor that can handle multiple patients and set up a scheduling program where a doctor can only handle 16 patients during an 8 hr workday. Recipe Creator and Manager - Create a recipe class with ingredients and put them in a recipe manager program that organizes them into categories like desserts, main courses, or by ingredients like chicken, beef, soups, pies, etc. Image Gallery - Create an image abstract class and then a class that inherits from it for each image type. Put them in a program that displays them in a gallery-style format for viewing. Shape Area and Perimeter Classes - Create an abstract class called Shape and then inherit from it other shapes like diamond, rectangle, circle, triangle, etc. Then have each class override the area and perimeter functionality to handle each shape type. Flower Shop Ordering To Go - Create a flower shop application that deals in flower objects and use those flower objects in a bouquet object which can then be sold. Keep track of the number of objects and when you may need to order more. Family Tree Creator - Create a class called Person which will have a name, when they were born, and when (and if) they died. Allow the user to create these Person classes and put them into a family tree structure. Print out the tree to the screen. Threading Create A Progress Bar for Downloads - Create a progress bar for applications that can keep track of a download in progress. The progress bar will be on a separate thread and will communicate with the main thread using delegates. Bulk Thumbnail Creator - Picture processing can take a bit of time for some transformations. Especially if the image is large. Create an image program that can take hundreds of images and converts them to a specified size in the background thread while you do other things. For added complexity, have one thread handling re-sizing, have another bulk renaming of thumbnails, etc. Web Page Scraper - Create an application that connects to a site and pulls out all links, or images, and saves them to a list. Optional: Organize the indexed content and don’t allow duplicates. Have it put the results into an easily searchable index file. Online White Board - Create an application that allows you to draw pictures, write notes and use various colors to flesh out ideas for projects. Optional: Add a feature to invite friends to collaborate on a whiteboard online. Get Atomic Time from Internet Clock - This program will get the true atomic time from an atomic time clock on the Internet. Use any one of the atomic clocks returned by a simple Google search. Fetch Current Weather - Get the current weather for a given zip/postal code. Optional: Try locating the user automatically. Scheduled Auto Login and Action - Make an application that logs into a given site on a schedule and invokes a certain action and then logs out. This can be useful for checking webmail, posting regular content, or getting info for other applications and saving it to your computer. E-Card Generator - Make a site that allows people to generate their own little e-cards and send them to other people. Do not use Flash. Use a picture library and perhaps insightful mottos or quotes. Content Management System - Create a content management system (CMS) like Joomla, Drupal, PHP Nuke, etc. Start small. Optional: Allow for the addition of modules/addons. Web Board (Forum) - Create a forum for you and your buddies to post, administer and share thoughts and ideas. CAPTCHA Maker - Ever see those images with letters numbers when you signup for a service and then ask you to enter what you see? It keeps web bots from automatically signing up and spamming. Try creating one yourself for online forms. Files Quiz Maker - Make an application that takes various questions from a file, picked randomly, and puts together a quiz for students. Each quiz can be different and then reads a key to grade the quizzes. Sort Excel/CSV File Utility - Reads a file of records, sorts them, and then writes them back to the file. Allow the user to choose various sort style and sorting based on a particular field. Create Zip File Maker - The user enters various files from different directories and the program zips them up into a zip file. Optional: Apply actual compression to the files. Start with Huffman Algorithm. PDF Generator - An application that can read in a text file, HTML file, or some other file and generates a PDF file out of it. Great for a web-based service where the user uploads the file and the program returns a PDF of the file. Optional: Deploy on GAE or Heroku if possible. Mp3 Tagger - Modify and add ID3v1 tags to MP3 files. See if you can also add in the album art into the MP3 file’s header as well as other ID3v2 tags. Code Snippet Manager - Another utility program that allows coders to put in functions, classes, or other tidbits to save for use later. Organized by the type of snippet or language the coder can quickly lookup code. Optional: For extra practice try adding syntax highlighting based on the language. Databases SQL Query Analyzer - A utility application in which a user can enter a query and have it run against a local database and look for ways to make it more efficient. Remote SQL Tool - A utility that can execute queries on remote servers from your local computer across the Internet. It should take in a remote host, user name, and password, run the query and return the results. Report Generator - Create a utility that generates a report based on some tables in a database. Generates sales reports based on the order/order details tables or sums up the day's current database activity. Event Scheduler and Calendar - Make an application that allows the user to enter a date and time of an event, event notes, and then schedule those events on a calendar. The user can then browse the calendar or search the calendar for specific events. Optional: Allow the application to create re-occurrence events that reoccur every day, week, month, year, etc. Budget Tracker - Write an application that keeps track of a household’s budget. The user can add expenses, income, and recurring costs to find out how much they are saving or losing over a period of time. Optional: Allow the user to specify a date range and see the net flow of money in and out of the house budget for that time period. TV Show Tracker - Got a favorite show you don’t want to miss? Don’t have a PVR or want to be able to find the show to then PVR it later? Make an application that can search various online TV Guide sites, locate the shows/times/channels and add them to a database application. The database/website then can send you email reminders that a show is about to start and which channel it will be on. Travel Planner System - Make a system that allows users to put together their own little travel itinerary and keep track of the airline/hotel arrangements, points of interest, budget, and schedule. Graphics and Multimedia Slide Show - Make an application that shows various pictures in a slide show format. Optional: Try adding various effects like fade in/out, star wipe, and window blinds transitions. Stream Video from Online - Try to create your own online streaming video player. Mp3 Player - A simple program for playing your favorite music files. Add features you think are missing from your favorite music player. Watermarking Application - Have some pictures you want copyright protected? Add your own logo or text lightly across the background so that no one can simply steal your graphics off your site. Make a program that will add this watermark to the picture. Optional: Use threading to process multiple images simultaneously. Turtle Graphics - This is a common project where you create a floor of 20 x 20 squares. Using various commands you tell a turtle to draw a line on the floor. You have moved forward, left or right, lift or drop the pen, etc. Do a search online for "Turtle Graphics" for more information. Optional: Allow the program to read in the list of commands from a file. GIF Creator A program that puts together multiple images (PNGs, JPGs, TIFFs) to make a smooth GIF that can be exported. Optional: Make the program convert small video files to GIFs as well. Security Caesar cipher - Implement a Caesar cipher, both encoding, and decoding. The key is an integer from 1 to 25. This cipher rotates the letters of the alphabet (A to Z). The encoding replaces each letter with the 1st to 25th next letter in the alphabet (wrapping Z to A). So key 2 encrypts "HI" to "JK", but key 20 encrypts "HI" to "BC". This simple "monoalphabetic substitution cipher" provides almost no security, because an attacker who has the encoded message can either use frequency analysis to guess the key, or just try all 25 keys.
alecwangcq / EigenDamage PytorchCode for "EigenDamage: Structured Pruning in the Kronecker-Factored Eigenbasis" https://arxiv.org/abs/1905.05934
MinhasKamal / AlgorithmImplementationsImplementation of Elementary Algorithms (infix-prefix-postfix-evaluation-to-longest-common-increasing-sub-sequence-activity-selection-balance-kd-binary-heap-binomial-tree-breath-depth-first-search-max-flow-shortest-path-topological-sort-calculus-derivative-integration-forward-interpolation-simpson-rule-intersecting-area-non-linear-equation-jacobis-gauss-seidal-bisection-false-position-newton-raphson-fixed-point-secant-cigarette-smokers-genetic-huffman-a-a*-star-binary-knuth-morris-pratt-kmp-quick-thread-priority-based-premitive-shortest-job-non-primitive-arithmetic-expression-data-structures-list-node-implementation-one-two-way-linked-stack-string-graph-numerical-methods-equation-solving-solve-process-problem-search-sort-prime-ugly-friend-perfect-fibonacci-factorial-factor-number)
hjeffreywang / Stock Feature EngineeringCreated a continuous, homogeneous, and structured 10 GB dataset from self obtained collections of unstructured intraday financial data. Generated features from indicators, statistics, and recent factors. Used multi-disciplined analysis to find feature importance. Attached labels of trends and stop/hold positions for machine learning. Used machine learning to significant features.
yanyan-li / Open StructureThis new benchmark dataset, Open-Structure, is proposed to evaluate visual odometry and SLAM methods, which directly equips point and line measurements, correspondences, structural associations, and co-visibility factor graphs instead of providing raw images.
Elijas / Baml AgentsBuilding Agents with LLM structured generation (BAML), MCP Tools, and 12-Factor Agents principles
ShelvanLee / XFEM# XFEM_Fracture2D ### Description This is a Matlab program that can be used to solve fracture problems involving arbitrary multiple crack propagations in a 2D linear-elastic solid based on the principle of minimum potential energy. The extended finite element method is used to discretise the solid continuum considering cracks as discontinuities in the displacement field. To this end, a strong discontinuity enrichment and a square-root singular crack tip enrichment are used to describe each crack. Several crack growth criteria are available to determine the evolution of cracks over time; apart from the classic maximum tension (or hoop-stress) criterion, the minimum total energy criterion and the local symmetry criterion are implemented implicitly with respect to the discrete time-stepping. ### Key features * *Fast:* The stiffness matrix and the force vector (i.e. the equations' system) and the enrichment tracking data structures are updated at each time step only with respect to the changes in the fracture topology. This ultimately results in the major part of the computational expense in the solution to the linear system of equations rather than in the post-processing of the solution or in the assembly and updating of the equations. As Matlab offers fast and robust direct solvers, the computational times are reasonably fast. * *Robust.* Suitable for multiple crack propagations with intersections. Furthermore, the stress intensity factors are computed robustly via the interaction integral approach (with the inclusion of the terms to account for crack surface pressure, residual stresses or strains). The minimum total energy criterion and the principle of local symmetry are implemented implicitly in time. The energy release rates are computed based on the stiffness derivative approach using algebraic differentiation (rather than finite differencing of the potential energy). On the other hand, the crack growth direction based on the local symmetry criterion is determined such that the local mode-II stress intensity factor vanishes; the change in a crack tip kink angle is approximated using the ratio of the crack tip stress intensity factors. * *Easy to run.* Each job has its own input files which are independent form those of all other jobs. The code especially lends itself to running parametric studies. Various results can be saved relating to the fracture geometry, fracture mechanics parameters, and the elastic fields in the solid domain. Extensive visualisation library is available for plotting results. ### Instructions 1. Get started by running the demo to showcase some of the capabilities of the program and to determine if it can be useful for you. At the Matlab's command line enter: ```Matlab >> RUN_JOBS.m ``` This will execute a series of jobs located inside the *jobs directory* `./JOBS_LIBRARY/`. These jobs do not take very long to execute (around 5 minutes in total). 2. Subsequently, you can pick one of the jobs inside `./JOBS_LIBRARY/` by defining the job title: ```Matlab >> job_title = 'several_cracks/edge/vertical_tension' ``` 3. Then you can open all the relevant scripts for this job as follows: ```Matlab >> open_job ``` The following input scripts for the *job* will be open in the Matlab's editor: 1. `JOB_MAIN.m`: This is the job's main script. It is called when executing `RUN_JOB` (or `RUN_JOBS`) and acts like a wrapper. Notably, it can serve as a convenient interface to run parametric studies and to save intermediate simulation results. 2. `Input_Scope.m`: This defines the scope of the simulation. From which crack growth criteria to use, to what to compute and what results to show via plots and/or movies. To put it simply, the script is a bunch of "switches" that tell the program what the user wants to be done. 3. `Input_Material.m`: Defines the material's elastic properties in different regions or layers (called "phases") of the computational domain. Moreover, it defines the fracture toughness of the material (assumed to be constant in all material phases). 4. `Input_Crack.m`: Defines the initial crack geometry. 5. `Input_BC.m`: Defines boundary conditions, such as displacements, tractions, crack surface pressure (assumed to be constant in all cracks), body loads (e.g. gravity, pre-stress or pre-strain). 6. `Mesh_make.m`: In-house structured mesh generator for rectangular domains using either linear triangle or bilinear quadrilateral elements. It is possible to mesh horizontal layers using different mesh sizes. 7. `Mesh_read.m`: Gmsh based mesh reader for version-1 mesh files. Of course you can use your own mesh reader provided the output variables are of the correct format (see later). 8. `Mesh_file.m`: Specifies the mesh input file (.msh). At the moment, only Gmsh mesh files of version-1 are allowed. ### Mesh_file.m A mesh file needs to be able to output the following data or variables: * `mNdCrd`: Node coordinates, size = `[nNdStd, 2]` * `mLNodS`: Element connectivities, size = `[nElemn,nLNodS]` * `vElPhz`: Element material phase (or region) ID's, size = `[nElemn,1]` * `cBCNod`: cell of boundary nodes, cell size = `{nBound,1}`, cell element size = `[nBnNod,2]` Example mesh files are located in `./JOBS_LIBRARY/`. Gmsh version-1 file format is described [here](http://www.manpagez.com/info/gmsh/gmsh-2.4.0/gmsh_60.php). ### Additional notes * global variables are defined in `.\Routines_AuxInput\Declare_Global.m` * External libraries are `.\Other_Libs\distmesh` and `.\Other_Libs\mesh2d` ### References Two external meshing libraries are used for the local mesh refinement and remeshing at the crack tip during crack propagation or prior to a crack intersection with another crack or with a boundary of the domain. Specifically, these libraries, which are located in `.\Other_Libs\`, are the following: * [*mesh2d*](https://people.sc.fsu.edu/~jburkardt/m_src/mesh2d/mesh2d.html) by Darren Engwirda * [*distmesh*](http://persson.berkeley.edu/distmesh/) by Per-Olof Persson and Gilbert Strang. ### Issues and Support For support or questions please email [sutula.danas@gmail.com](mailto:sutula.danas@gmail.com). ### Authors Danas Sutula, University of Luxembourg, Luxembourg. If you find this code useful, we kindly ask that you consider citing us. * [Minimum energy multiple crack propagation](http://hdl.handle.net/10993/29414)
alefunxo / BasopraBASOPRA - BAttery Schedule OPtimizer for Residential Applications. Daily battery schedule optimizer (i.e. 24 h optimization framework), assuming perfect day-ahead forecast of the electricity demand load and solar PV generation in order to determine the maximum economic potential regardless of the forecast strategy used. Include the use of different applications which residential batteries can perform from a consumer perspective. Applications such as avoidance of PV curtailment, demand load-shifting and demand peak shaving are considered along with the base application, PV self-consumption. Different battery technologies and sizes can be analyzed as well as different tariff structures. Aging is treated as an exogenous parameter, calculated on daily basis and is not subject of optimization. Data with 15-minute temporal resolution are used for simulations. The model objective function have two components, the energy-based and the power-based component, as the tariff structure depends on the applications considered, a boolean parameter activate the power-based factor of the bill when is necessary.
HuysmanWang / Deep Learning Aided Porous Media Hydrodynamic Analysis And Three Dimensional ReconstructionThe study of hydrodynamic behavior and water-rock interaction mechanisms is typically characterized by high computational efficiency requirements, to allow for the fast and accurate extraction of structural information. Therefore, we chose to use deep learning models to achieve these requirements. In this paper we started by comparing the image segmentation performance of a series of autoencoder architectures on complex geometries of porous media. The goal was to extract hydrodynamic connectivity channels and the mineral composition of rock samples on SEM (Scanning electron microscopy) data, obtained with a 0.97 accuracy. We then focused on improving the computational efficiency of LBM by using GPU acceleration, which allowed us to rapidly simulate structural flow field features of complex porous media. The results obtained showed that we were able to improve the computational efficiency by a factor of 21 in our device environment. We subsequently employed a SWD-Cycle-GAN technique to migrate sedimentation features to the initial 2D structure slices to reconstruct a 3D (three-dimensional) porous media geometry, that fits the depositional features more closely. Overall, we propose a new method for 3D structure reconstruction and permeability performance analysis of porous media, based on deep learning. The proposed method is fast, efficient and accurate.
deep-spin / Lp SparsemapLP-SparseMAP: Differentiable sparse structured prediction in coarse factor graphs
anandtripathi5 / BarahBarah is a Python Twelve factor Flask boilerplate directory structure. You can use this 12 factor app as a template project for any python project that you want to build on flask.
AlfredCYL / Gplearn Cross FactorEnhance the gplearn package to support precise three-dimensional structured dimension genetic programming (GP), with a particular focus on enabling cross-sectional factor analysis within the package.
MrBriit / FLASK End To End Zomato Restaurant Price Prediction And Deployment# **ABSTRACT** Main Objective: The main agenda of this project is: Perform extensive Exploratory Data Analysis(EDA) on the Zomato Dataset. Build an appropriate Machine Learning Model that will help various Zomato Restaurants to predict their respective Ratings based on certain features DEPLOY the Machine learning model via Flask that can be used to make live predictions of restaurants ratings A step by step guide is attached to this documnet as well as a video explanation of each concpet. Zomato is one of the best online food delivery apps which gives the users the ratings and the reviews on restaurants all over india.These ratings and the Reviews are considered as one of the most important deciding factors which determine how good a restaurant is. We will therefore use the real time Data set with variuos features a user would look into regarding a restaurant. We will be considering Banglore City in this analysis. Content The basic idea of analyzing the Zomato dataset is to get a fair idea about the factors affecting the establishment of different types of restaurant at different places in Bengaluru, aggregate rating of each restaurant, Bengaluru being one such city has more than 12,000 restaurants with restaurants serving dishes from all over the world. With each day new restaurants opening the industry has’nt been saturated yet and the demand is increasing day by day. Inspite of increasing demand it however has become difficult for new restaurants to compete with established restaurants. Most of them serving the same food. Bengaluru being an IT capital of India. Most of the people here are dependent mainly on the restaurant food as they don’t have time to cook for themselves. With such an overwhelming demand of restaurants it has therefore become important to study the demography of a location. What kind of a food is more popular in a locality. Do the entire locality loves vegetarian food. If yes then is that locality populated by a particular sect of people for eg. Jain, Marwaris, Gujaratis who are mostly vegetarian. These kind of analysis can be done using the data, by studying the factors such as • Location of the restaurant • Approx Price of food • Theme based restaurant or not • Which locality of that city serves that cuisines with maximum number of restaurants • The needs of people who are striving to get the best cuisine of the neighborhood • Is a particular neighborhood famous for its own kind of food. “Just so that you have a good meal the next time you step out” The data is accurate to that available on the zomato website until 15 March 2019. The data was scraped from Zomato in two phase. After going through the structure of the website I found that for each neighborhood there are 6-7 category of restaurants viz. Buffet, Cafes, Delivery, Desserts, Dine-out, Drinks & nightlife, Pubs and bars. Phase I, In Phase I of extraction only the URL, name and address of the restaurant were extracted which were visible on the front page. The URl's for each of the restaurants on the zomato were recorded in the csv file so that later the data can be extracted individually for each restaurant. This made the extraction process easier and reduced the extra load on my machine. The data for each neighborhood and each category can be found here Phase II, In Phase II the recorded data for each restaurant and each category was read and data for each restaurant was scraped individually. 15 variables were scraped in this phase. For each of the neighborhood and for each category their onlineorder, booktable, rate, votes, phone, location, resttype, dishliked, cuisines, approxcost(for two people), reviewslist, menu_item was extracted. See section 5 for more details about the variables. Acknowledgements The data scraped was entirely for educational purposes only. Note that I don’t claim any copyright for the data. All copyrights for the data is owned by Zomato Media Pvt. Ltd.. Source: Kaggle
sayantann11 / Clustering Modelsfor MLlustering in Machine Learning Introduction to Clustering It is basically a type of unsupervised learning method . An unsupervised learning method is a method in which we draw references from datasets consisting of input data without labelled responses. Generally, it is used as a process to find meaningful structure, explanatory underlying processes, generative features, and groupings inherent in a set of examples. Clustering is the task of dividing the population or data points into a number of groups such that data points in the same groups are more similar to other data points in the same group and dissimilar to the data points in other groups. It is basically a collection of objects on the basis of similarity and dissimilarity between them. For ex– The data points in the graph below clustered together can be classified into one single group. We can distinguish the clusters, and we can identify that there are 3 clusters in the below picture. It is not necessary for clusters to be a spherical. Such as : DBSCAN: Density-based Spatial Clustering of Applications with Noise These data points are clustered by using the basic concept that the data point lies within the given constraint from the cluster centre. Various distance methods and techniques are used for calculation of the outliers. Why Clustering ? Clustering is very much important as it determines the intrinsic grouping among the unlabeled data present. There are no criteria for a good clustering. It depends on the user, what is the criteria they may use which satisfy their need. For instance, we could be interested in finding representatives for homogeneous groups (data reduction), in finding “natural clusters” and describe their unknown properties (“natural” data types), in finding useful and suitable groupings (“useful” data classes) or in finding unusual data objects (outlier detection). This algorithm must make some assumptions which constitute the similarity of points and each assumption make different and equally valid clusters. Clustering Methods : Density-Based Methods : These methods consider the clusters as the dense region having some similarity and different from the lower dense region of the space. These methods have good accuracy and ability to merge two clusters.Example DBSCAN (Density-Based Spatial Clustering of Applications with Noise) , OPTICS (Ordering Points to Identify Clustering Structure) etc. Hierarchical Based Methods : The clusters formed in this method forms a tree-type structure based on the hierarchy. New clusters are formed using the previously formed one. It is divided into two category Agglomerative (bottom up approach) Divisive (top down approach) examples CURE (Clustering Using Representatives), BIRCH (Balanced Iterative Reducing Clustering and using Hierarchies) etc. Partitioning Methods : These methods partition the objects into k clusters and each partition forms one cluster. This method is used to optimize an objective criterion similarity function such as when the distance is a major parameter example K-means, CLARANS (Clustering Large Applications based upon Randomized Search) etc. Grid-based Methods : In this method the data space is formulated into a finite number of cells that form a grid-like structure. All the clustering operation done on these grids are fast and independent of the number of data objects example STING (Statistical Information Grid), wave cluster, CLIQUE (CLustering In Quest) etc. Clustering Algorithms : K-means clustering algorithm – It is the simplest unsupervised learning algorithm that solves clustering problem.K-means algorithm partition n observations into k clusters where each observation belongs to the cluster with the nearest mean serving as a prototype of the cluster . Applications of Clustering in different fields Marketing : It can be used to characterize & discover customer segments for marketing purposes. Biology : It can be used for classification among different species of plants and animals. Libraries : It is used in clustering different books on the basis of topics and information. Insurance : It is used to acknowledge the customers, their policies and identifying the frauds. City Planning: It is used to make groups of houses and to study their values based on their geographical locations and other factors present. Earthquake studies: By learning the earthquake-affected areas we can determine the dangerous zones. References : Wiki Hierarchical clustering Ijarcs matteucc analyticsvidhya knowm
Mdshobu / Liberty House Club Whitepaper# Liberty House Club **A Parallel Binance Chain to Enable Smart Contracts** _NOTE: This document is under development. Please check regularly for updates!_ ## Table of Contents - [Motivation](#motivation) - [Design Principles](#design-principles) - [Consensus and Validator Quorum](#consensus-and-validator-quorum) * [Proof of Staked Authority](#proof-of-staked-authority) * [Validator Quorum](#validator-quorum) * [Security and Finality](#security-and-finality) * [Reward](#reward) - [Token Economy](#token-economy) * [Native Token](#native-token) * [Other Tokens](#other-tokens) - [Cross-Chain Transfer and Communication](#cross-chain-transfer-and-communication) * [Cross-Chain Transfer](#cross-chain-transfer) * [BC to BSC Architecture](#bc-to-bsc-architecture) * [BSC to BC Architecture](#bsc-to-bc-architecture) * [Timeout and Error Handling](#timeout-and-error-handling) * [Cross-Chain User Experience](#cross-chain-user-experience) * [Cross-Chain Contract Event](#cross-chain-contract-event) - [Staking and Governance](#staking-and-governance) * [Staking on BC](#staking-on-bc) * [Rewarding](#rewarding) * [Slashing](#slashing) - [Relayers](#relayers) * [BSC Relayers](#bsc-relayers) * [Oracle Relayers](#oracle-relayers) - [Outlook](#outlook) # Motivation After its mainnet community [launch](https://www.binance.com/en/blog/327334696200323072/Binance-DEX-Launches-on-Binance-Chain-Invites-Further-Community-Development) in April 2019, [Binance Chain](https://www.binance.org) has exhibited its high speed and large throughput design. Binance Chain’s primary focus, its native [decentralized application](https://en.wikipedia.org/wiki/Decentralized_application) (“dApp”) [Binance DEX](https://www.binance.org/trade), has demonstrated its low-latency matching with large capacity headroom by handling millions of trading volume in a short time. Flexibility and usability are often in an inverse relationship with performance. The concentration on providing a convenient digital asset issuing and trading venue also brings limitations. Binance Chain's most requested feature is the programmable extendibility, or simply the [Smart Contract](https://en.wikipedia.org/wiki/Smart_contract) and Virtual Machine functions. Digital asset issuers and owners struggle to add new decentralized features for their assets or introduce any sort of community governance and activities. Despite this high demand for adding the Smart Contract feature onto Binance Chain, it is a hard decision to make. The execution of a Smart Contract may slow down the exchange function and add non-deterministic factors to trading. If that compromise could be tolerated, it might be a straightforward idea to introduce a new Virtual Machine specification based on [Tendermint](https://tendermint.com/core/), based on the current underlying consensus protocol and major [RPC](https://docs.binance.org/api-reference/node-rpc.html) implementation of Binance Chain. But all these will increase the learning requirements for all existing dApp communities, and will not be very welcomed. We propose a parallel blockchain of the current Binance Chain to retain the high performance of the native DEX blockchain and to support a friendly Smart Contract function at the same time. # Design Principles After the creation of the parallel blockchain into the Binance Chain ecosystem, two blockchains will run side by side to provide different services. The new parallel chain will be called “**Binance Smart Chain**” (short as “**BSC**” for the below sections), while the existing mainnet remains named “**Binance Chain**” (short as “**BC**” for the below sections). Here are the design principles of **BSC**: 1. **Standalone Blockchain**: technically, BSC is a standalone blockchain, instead of a layer-2 solution. Most BSC fundamental technical and business functions should be self-contained so that it can run well even if the BC stopped for a short period. 2. **Ethereum Compatibility**: The first practical and widely-used Smart Contract platform is Ethereum. To take advantage of the relatively mature applications and community, BSC chooses to be compatible with the existing Ethereum mainnet. This means most of the **dApps**, ecosystem components, and toolings will work with BSC and require zero or minimum changes; BSC node will require similar (or a bit higher) hardware specification and skills to run and operate. The implementation should leave room for BSC to catch up with further Ethereum upgrades. 3. **Staking Involved Consensus and Governance**: Staking-based consensus is more environmentally friendly and leaves more flexible option to the community governance. Expectedly, this consensus should enable better network performance over [proof-of-work](https://en.wikipedia.org/wiki/Proof_of_work) blockchain system, i.e., faster blocking time and higher transaction capacity. 4. **Native Cross-Chain Communication**: both BC and BSC will be implemented with native support for cross-chain communication among the two blockchains. The communication protocol should be bi-directional, decentralized, and trustless. It will concentrate on moving digital assets between BC and BSC, i.e., [BEP2](https://github.com/binance-chain/BEPs/blob/master/BEP2.md) tokens, and eventually, other BEP tokens introduced later. The protocol should care for the minimum of other items stored in the state of the blockchains, with only a few exceptions. # Consensus and Validator Quorum Based on the above design principles, the consensus protocol of BSC is to fulfill the following goals: 1. Blocking time should be shorter than Ethereum network, e.g. 5 seconds or even shorter. 2. It requires limited time to confirm the finality of transactions, e.g. around 1-min level or shorter. 3. There is no inflation of native token: BNB, the block reward is collected from transaction fees, and it will be paid in BNB. 4. It is compatible with Ethereum system as much as possible. 5. It allows modern [proof-of-stake](https://en.wikipedia.org/wiki/Proof_of_stake) blockchain network governance. ## Proof of Staked Authority Although Proof-of-Work (PoW) has been recognized as a practical mechanism to implement a decentralized network, it is not friendly to the environment and also requires a large size of participants to maintain the security. Ethereum and some other blockchain networks, such as [MATIC Bor](https://github.com/maticnetwork/bor), [TOMOChain](https://tomochain.com/), [GoChain](https://gochain.io/), [xDAI](https://xdai.io/), do use [Proof-of-Authority(PoA)](https://en.wikipedia.org/wiki/Proof_of_authority) or its variants in different scenarios, including both testnet and mainnet. PoA provides some defense to 51% attack, with improved efficiency and tolerance to certain levels of Byzantine players (malicious or hacked). It serves as an easy choice to pick as the fundamentals. Meanwhile, the PoA protocol is most criticized for being not as decentralized as PoW, as the validators, i.e. the nodes that take turns to produce blocks, have all the authorities and are prone to corruption and security attacks. Other blockchains, such as EOS and Lisk both, introduce different types of [Delegated Proof of Stake (DPoS)](https://en.bitcoinwiki.org/wiki/DPoS) to allow the token holders to vote and elect the validator set. It increases the decentralization and favors community governance. BSC here proposes to combine DPoS and PoA for consensus, so that: 1. Blocks are produced by a limited set of validators 2. Validators take turns to produce blocks in a PoA manner, similar to [Ethereum’s Clique](https://eips.ethereum.org/EIPS/eip-225) consensus design 3. Validator set are elected in and out based on a staking based governance ## Validator Quorum In the genesis stage, a few trusted nodes will run as the initial Validator Set. After the blocking starts, anyone can compete to join as candidates to elect as a validator. The staking status decides the top 21 most staked nodes to be the next validator set, and such an election will repeat every 24 hours. **BNB** is the token used to stake for BSC. In order to remain as compatible as Ethereum and upgradeable to future consensus protocols to be developed, BSC chooses to rely on the **BC** for staking management (Please refer to the below “[Staking and Governance](#staking-and-governance)” section). There is a **dedicated staking module for BSC on BC**. It will accept BSC staking from BNB holders and calculate the highest staked node set. Upon every UTC midnight, BC will issue a verifiable `ValidatorSetUpdate` cross-chain message to notify BSC to update its validator set. While producing further blocks, the existing BSC validators check whether there is a `ValidatorSetUpdate` message relayed onto BSC periodically. If there is, they will update the validator set after an **epoch period**, i.e. a predefined number of blocking time. For example, if BSC produces a block every 5 seconds, and the epoch period is 240 blocks, then the current validator set will check and update the validator set for the next epoch in 1200 seconds (20 minutes). ## Security and Finality Given there are more than ½\*N+1 validators are honest, PoA based networks usually work securely and properly. However, there are still cases where certain amount Byzantine validators may still manage to attack the network, e.g. through the “[Clone Attack](https://arxiv.org/pdf/1902.10244.pdf)”. To secure as much as BC, BSC users are encouraged to wait until receiving blocks sealed by more than ⅔\*N+1 different validators. In that way, the BSC can be trusted at a similar security level to BC and can tolerate less than ⅓\*N Byzantine validators. With 21 validators, if the block time is 5 seconds, the ⅔\*N+1 different validator seals will need a time period of (⅔\*21+1)*5 = 75 seconds. Any critical applications for BSC may have to wait for ⅔\*N+1 to ensure a relatively secure finality. However, besides such arrangement, BSC does introduce **Slashing** logic to penalize Byzantine validators for **double signing** or **inavailability**, which will be covered in the “Staking and Governance” section later. This Slashing logic will expose the malicious validators in a very short time and make the “Clone Attack” very hard or extremely non-beneficial to execute. With this enhancement, ½\*N+1 or even fewer blocks are enough as confirmation for most transactions. ## Reward All the BSC validators in the current validator set will be rewarded with transaction **fees in BNB**. As BNB is not an inflationary token, there will be no mining rewards as what Bitcoin and Ethereum network generate, and the gas fee is the major reward for validators. As BNB is also utility tokens with other use cases, delegators and validators will still enjoy other benefits of holding BNB. The reward for validators is the fees collected from transactions in each block. Validators can decide how much to give back to the delegators who stake their BNB to them, in order to attract more staking. Every validator will take turns to produce the blocks in the same probability (if they stick to 100% liveness), thus, in the long run, all the stable validators may get a similar size of the reward. Meanwhile, the stakes on each validator may be different, so this brings a counter-intuitive situation that more users trust and delegate to one validator, they potentially get less reward. So rational delegators will tend to delegate to the one with fewer stakes as long as the validator is still trustful (insecure validator may bring slashable risk). In the end, the stakes on all the validators will have less variation. This will actually prevent the stake concentration and “winner wins forever” problem seen on some other networks. Some parts of the gas fee will also be rewarded to relayers for Cross-Chain communication. Please refer to the “[Relayers](#relayers)” section below. # Token Economy BC and BSC share the same token universe for BNB and BEP2 tokens. This defines: 1. The same token can circulate on both networks, and flow between them bi-directionally via a cross-chain communication mechanism. 2. The total circulation of the same token should be managed across the two networks, i.e. the total effective supply of a token should be the sum of the token’s total effective supply on both BSC and BC. 3. The tokens can be initially created on BSC in a similar format as ERC20 token standard, or on BC as a BEP2, then created on the other. There are native ways on both networks to link the two and secure the total supply of the token. ## Native Token BNB will run on BSC in the same way as ETH runs on Ethereum so that it remains as “native token” for both BSC and BC. This means, in addition to BNB is used to pay most of the fees on Binance Chain and Binance DEX, BNB will be also used to: 1. pay “fees“ to deploy smart contracts on BSC 2. stake on selected BSC validators, and get corresponding rewards 3. perform cross-chain operations, such as transfer token assets across BC and BSC ### Seed Fund Certain amounts of BNB will be burnt on BC and minted on BSC during its genesis stage. This amount is called “Seed Fund” to circulate on BSC after the first block, which will be dispatched to the initial BC-to-BSC Relayer(described in later sections) and initial validator set introduced at genesis. These BNBs are used to pay transaction fees in the early stage to transfer more BNB from BC onto BSC via the cross-chain mechanism. The BNB cross-chain transfer is discussed in a later section, but for BC to BSC transfer, it is generally to lock BNB on BC from the source address of the transfer to a system-controlled address and unlock the corresponding amount from special contract to the target address of the transfer on BSC, or reversely, when transferring from BSC to BC, it is to lock BNB from the source address on BSC into a special contract and release locked amount on BC from the system address to the target address. The logic is related to native code on BC and a series of smart contracts on BSC. ## Other Tokens BC supports BEP2 tokens and upcoming [BEP8 tokens](https://github.com/binance-chain/BEPs/pull/69), which are native assets transferrable and tradable (if listed) via fast transactions and sub-second finality. Meanwhile, as BSC is Ethereum compatible, it is natural to support ERC20 tokens on BSC, which here is called “**BEP2E**” (with the real name to be introduced by the future BEPs,it potentially covers BEP8 as well). BEP2E may be “Enhanced” by adding a few more methods to expose more information, such as token denomination, decimal precision definition and the owner address who can decide the Token Binding across the chains. BSC and BC work together to ensure that one token can circulate in both formats with confirmed total supply and be used in different use cases. ### Token Binding BEP2 tokens will be extended to host a new attribute to associate the token with a BSC BEP2E token contract, called “**Binder**”, and this process of association is called “**Token Binding**”. Token Binding can happen at any time after BEP2 and BEP2E are ready. The token owners of either BEP2 or BEP2E don’t need to bother about the Binding, until before they really want to use the tokens on different scenarios. Issuers can either create BEP2 first or BEP2E first, and they can be bound at a later time. Of course, it is encouraged for all the issuers of BEP2 and BEP2E to set the Binding up early after the issuance. A typical procedure to bind the BEP2 and BEP2E will be like the below: 1. Ensure both the BEP2 token and the BEP2E token both exist on each blockchain, with the same total supply. BEP2E should have 3 more methods than typical ERC20 token standard: * symbol(): get token symbol * decimals(): get the number of the token decimal digits * owner(): get **BEP2E contract owner’s address.** This value should be initialized in the BEP2E contract constructor so that the further binding action can verify whether the action is from the BEP2E owner. 2. Decide the initial circulation on both blockchains. Suppose the total supply is *S*, and the expected initial circulating supply on BC is *K*, then the owner should lock S-K tokens to a system controlled address on BC. 3. Equivalently, *K* tokens is locked in the special contract on BSC, which handles major binding functions and is named as **TokenHub**. The issuer of the BEP2E token should lock the *K* amount of that token into TokenHub, resulting in *S-K* tokens to circulate on BSC. Thus the total circulation across 2 blockchains remains as *S*. 4. The issuer of BEP2 token sends the bind transaction on BC. Once the transaction is executed successfully after proper verification: * It transfers *S-K* tokens to a system-controlled address on BC. * A cross-chain bind request package will be created, waiting for Relayers to relay. 5. BSC Relayers will relay the cross-chain bind request package into **TokenHub** on BSC, and the corresponding request and information will be stored into the contract. 6. The contract owner and only the owner can run a special method of TokenHub contract, `ApproveBind`, to verify the binding request to mark it as a success. It will confirm: * the token has not been bound; * the binding is for the proper symbol, with proper total supply and decimal information; * the proper lock are done on both networks; 10. Once the `ApproveBind` method has succeeded, TokenHub will mark the two tokens are bounded and share the same circulation on BSC, and the status will be propagated back to BC. After this final confirmation, the BEP2E contract address and decimals will be written onto the BEP2 token as a new attribute on BC, and the tokens can be transferred across the two blockchains bidirectionally. If the ApproveBind fails, the failure event will also be propagated back to BC to release the locked tokens, and the above steps can be re-tried later. # Cross-Chain Transfer and Communication Cross-chain communication is the key foundation to allow the community to take advantage of the dual chain structure: * users are free to create any tokenization, financial products, and digital assets on BSC or BC as they wish * the items on BSC can be manually and programmingly traded and circulated in a stable, high throughput, lighting fast and friendly environment of BC * users can operate these in one UI and tooling ecosystem. ## Cross-Chain Transfer The cross-chain transfer is the key communication between the two blockchains. Essentially the logic is: 1. the `transfer-out` blockchain will lock the amount from source owner addresses into a system controlled address/contracts; 2. the `transfer-in` blockchain will unlock the amount from the system controlled address/contracts and send it to target addresses. The cross-chain transfer package message should allow the BSC Relayers and BC **Oracle Relayers** to verify: 1. Enough amount of token assets are removed from the source address and locked into a system controlled addresses/contracts on the source blockchain. And this can be confirmed on the target blockchain. 2. Proper amounts of token assets are released from a system controlled addresses/contracts and allocated into target addresses on the target blockchain. If this fails, it can be confirmed on source blockchain, so that the locked token can be released back (may deduct fees). 3. The sum of the total circulation of the token assets across the 2 blockchains are not changed after this transfer action completes, no matter if the transfer succeeds or not.  The architecture of cross-chain communication is as in the above diagram. To accommodate the 2 heteroid systems, communication handling is different in each direction. ## BC to BSC Architecture BC is a Tendermint-based, instant finality blockchain. Validators with at least ⅔\*N+1 of the total voting power will co-sign each block on the chain. So that it is practical to verify the block transactions and even the state value via **Block Header** and **Merkle Proof** verification. This has been researched and implemented as “**Light-Client Protocol**”, which are intensively discussed in [the Ethereum](https://github.com/ethereum/wiki/wiki/Light-client-protocol) community, studied and implemented for [Cosmos inter-chain communication](https://github.com/cosmos/ics/blob/a4173c91560567bdb7cc9abee8e61256fc3725e9/spec/ics-007-tendermint-client/README.md). BC-to-BSC communication will be verified in an “**on-chain light client**” implemented via BSC **Smart Contracts** (some of them may be **“pre-compiled”**). After some transactions and state change happen on BC, if a transaction is defined to trigger cross-chain communication,the Cross-chain “**package**” message will be created and **BSC Relayers** will pass and submit them onto BSC as data into the "build-in system contracts". The build-in system contracts will verify the package and execute the transactions if it passes the verification. The verification will be guaranteed with the below design: 1. BC blocking status will be synced to the light client contracts on BSC from time to time, via block header and pre-commits, for the below information: * block and app hash of BC that are signed by validators * current validatorset, and validator set update 2. the key-value from the blockchain state will be verified based on the Merkle Proof and information from above #1. After confirming the key-value is accurate and trustful, the build-in system contracts will execute the actions corresponding to the cross-chain packages. Some examples of such packages that can be created for BC-to-BSC are: 1. Bind: bind the BEP2 tokens and BEP2E 2. Transfer: transfer tokens after binding, this means the circulation will decrease (be locked) from BC and appear in the target address balance on BSC 3. Error Handling: to handle any timeout/failure event for BSC-to-BC communication 4. Validatorset update of BSC To ensure no duplication, proper message sequence and timely timeout, there is a “Channel” concept introduced on BC to manage any types of the communication. For relayers, please also refer to the below “Relayers” section. ## BSC to BC Architecture BSC uses Proof of Staked Authority consensus protocol, which has a chance to fork and requires confirmation of more blocks. One block only has the signature of one validator, so that it is not easy to rely on one block to verify data from BSC. To take full advantage of validator quorum of BC, an idea similar to many [Bridge ](https://github.com/poanetwork/poa-bridge)or Oracle blockchains is adopted: 1. The cross-chain communication requests from BSC will be submitted and executed onto BSC as transactions. The execution of the transanction wil emit `Events`, and such events can be observed and packaged in certain “**Oracle**” onto BC. Instead of Block Headers, Hash and Merkle Proof, this type of “Oracle” package directly contains the cross-chain information for actions, such as sender, receiver and amount for transfer. 2. To ensure the security of the Oracle, the validators of BC will form anothe quorum of “**Oracle Relayers**”. Each validator of the BC should run a **dedicated process** as the Oracle Relayer. These Oracle Relayers will submit and vote for the cross-chain communication package, like Oracle, onto BC, using the same validator keys. Any package signed by more than ⅔\*N+1 Oracle Relayers’ voting power is as secure as any block signed by ⅔\*N+1 of the same quorum of validators’ voting power. By using the same validator quorum, it saves the light client code on BC and continuous block updates onto BC. Such Oracles also have Oracle IDs and types, to ensure sequencing and proper error handling. ## Timeout and Error Handling There are scenarios that the cross-chain communication fails. For example, the relayed package cannot be executed on BSC due to some coding bug in the contracts. **Timeout and error handling logics are** used in such scenarios. For the recognizable user and system errors or any expected exceptions, the two networks should heal themselves. For example, when BC to BSC transfer fails, BSC will issue a failure event and Oracle Relayers will execute a refund on BC; when BSC to BC transfer fails, BC will issue a refund package for Relayer to relay in order to unlock the fund. However, unexpected error or exception may still happen on any step of the cross-chain communication. In such a case, the Relayers and Oracle Relayers will discover that the corresponding cross-chain channel is stuck in a particular sequence. After a Timeout period, the Relayers and Oracle Relayers can request a “SkipSequence” transaction, the stuck sequence will be marked as “Unexecutable”. A corresponding alerts will be raised, and the community has to discuss how to handle this scenario, e.g. payback via the sponsor of the validators, or event clear the fund during next network upgrade. ## Cross-Chain User Experience Ideally, users expect to use two parallel chains in the same way as they use one single chain. It requires more aggregated transaction types to be added onto the cross-chain communication to enable this, which will add great complexity, tight coupling, and maintenance burden. Here BC and BSC only implement the basic operations to enable the value flow in the initial launch and leave most of the user experience work to client side UI, such as wallets. E.g. a great wallet may allow users to sell a token directly from BSC onto BC’s DEX order book, in a secure way. ## Cross-Chain Contract Event Cross-Chain Contract Event (CCCE) is designed to allow a smart contract to trigger cross-chain transactions, directly through the contract code. This becomes possible based on: 1. Standard system contracts can be provided to serve operations callable by general smart contracts; 2. Standard events can be emitted by the standard contracts; 3. Oracle Relayers can capture the standard events, and trigger the corresponding cross-chain operations; 4. Dedicated, code-managed address (account) can be created on BC and accessed by the contracts on the BSC, here it is named as **“Contract Address on BC” (CAoB)**. Several standard operations are implemented: 1. BSC to BC transfer: this is implemented in the same way as normal BSC to BC transfer, by only triggered via standard contract. The fund can be transferred to any addresses on BC, including the corresponding CAoB of the transfer originating contract. 2. Transfer on BC: this is implemented as a special cross-chain transfer, while the real transfer is from **CAoB** to any other address (even another CAoB). 3. BC to BSC transfer: this is implemented as two-pass cross-chain communication. The first is triggered by the BSC contract and propagated onto BC, and then in the second pass, BC will start a normal BC to BSC cross-chain transfer, from **CAoB** to contract address on BSC. A special note should be paid on that the BSC contract only increases balance upon any transfer coming in on the second pass, and the error handling in the second pass is the same as the normal BC to BSC transfer. 4. IOC (Immediate-Or-Cancel) Trade Out: the primary goal of transferring assets to BC is to trade. This event will instruct to trade a certain amount of an asset in CAoB into another asset as much as possible and transfer out all the results, i.e. the left the source and the traded target tokens of the trade, back to BSC. BC will handle such relayed events by sending an “Immediate-Or-Cancel”, i.e. IOC order onto the trading pairs, once the next matching finishes, the result will be relayed back to BSC, which can be in either one or two assets. 5. Auction Trade Out: Such event will instruct BC to send an auction order to trade a certain amount of an asset in **CAoB** into another asset as much as possible and transfer out all the results back to BSC at the end of the auction. Auction function is upcoming on BC. There are some details for the Trade Out: 1. both can have a limit price (absolute or relative) for the trade; 2. the end result will be written as cross-chain packages to relay back to BSC; 3. cross-chain communication fees may be charged from the asset transferred back to BSC; 4. BSC contract maintains a mirror of the balance and outstanding orders on CAoB. No matter what error happens during the Trade Out, the final status will be propagated back to the originating contract and clear its internal state. With the above features, it simply adds the cross-chain transfer and exchange functions with high liquidity onto all the smart contracts on BSC. It will greatly add the application scenarios on Smart Contract and dApps, and make 1 chain +1 chain > 2 chains. # Staking and Governance Proof of Staked Authority brings in decentralization and community involvement. Its core logic can be summarized as the below. You may see similar ideas from other networks, especially Cosmos and EOS. 1. Token holders, including the validators, can put their tokens “**bonded**” into the stake. Token holders can **delegate** their tokens onto any validator or validator candidate, to expect it can become an actual validator, and later they can choose a different validator or candidate to **re-delegate** their tokens<sup>1</sup>. 2. All validator candidates will be ranked by the number of bonded tokens on them, and the top ones will become the real validators. 3. Validators can share (part of) their blocking reward with their delegators. 4. Validators can suffer from “**Slashing**”, a punishment for their bad behaviors, such as double sign and/or instability. 5. There is an “**unbonding period**” for validators and delegators so that the system makes sure the tokens remain bonded when bad behaviors are caught, the responsible will get slashed during this period. ## Staking on BC Ideally, such staking and reward logic should be built into the blockchain, and automatically executed as the blocking happens. Cosmos Hub, who shares the same Tendermint consensus and libraries with Binance Chain, works in this way. BC has been preparing to enable staking logic since the design days. On the other side, as BSC wants to remain compatible with Ethereum as much as possible, it is a great challenge and efforts to implement such logic on it. This is especially true when Ethereum itself may move into a different Proof of Stake consensus protocol in a short (or longer) time. In order to keep the compatibility and reuse the good foundation of BC, the staking logic of BSC is implemented on BC: 1. The staking token is BNB, as it is a native token on both blockchains anyway 2. The staking, i.e. token bond and delegation actions and records for BSC, happens on BC. 3. The BSC validator set is determined by its staking and delegation logic, via a staking module built on BC for BSC, and propagated every day UTC 00:00 from BC to BSC via Cross-Chain communication. 4. The reward distribution happens on BC around every day UTC 00:00. ## Rewarding Both the validator update and reward distribution happen every day around UTC 00:00. This is to save the cost of frequent staking updates and block reward distribution. This cost can be significant, as the blocking reward is collected on BSC and distributed on BC to BSC validators and delegators. (Please note BC blocking fees will remain rewarding to BC validators only.) A deliberate delay is introduced here to make sure the distribution is fair: 1. The blocking reward will not be sent to validator right away, instead, they will be distributed and accumulated on a contract; 2. Upon receiving the validator set update into BSC, it will trigger a few cross-chain transfers to transfer the reward to custody addresses on the corresponding validators. The custody addresses are owned by the system so that the reward cannot be spent until the promised distribution to delegators happens. 3. In order to make the synchronization simpler and allocate time to accommodate slashing, the reward for N day will be only distributed in N+2 days. After the delegators get the reward, the left will be transferred to validators’ own reward addresses. ## Slashing Slashing is part of the on-chain governance, to ensure the malicious or negative behaviors are punished. BSC slash can be submitted by anyone. The transaction submission requires **slash evidence** and cost fees but also brings a larger reward when it is successful. So far there are two slashable cases. ### Double Sign It is quite a serious error and very likely deliberate offense when a validator signs more than one block with the same height and parent block. The reference protocol implementation should already have logic to prevent this, so only the malicious code can trigger this. When Double Sign happens, the validator should be removed from the Validator **Set** right away. Anyone can submit a slash request on BC with the evidence of Double Sign of BSC, which should contain the 2 block headers with the same height and parent block, sealed by the offending validator. Upon receiving the evidence, if the BC verifies it to be valid: 1. The validator will be removed from validator set by an instance BSC validator set update Cross-Chain update; 2. A predefined amount of BNB would be slashed from the **self-delegated** BNB of the validator; Both validator and its delegators will not receive the staking rewards. 3. Part of the slashed BNB will allocate to the submitter’s address, which is a reward and larger than the cost of submitting slash request transaction 4. The rest of the slashed BNB will allocate to the other validators’ custody addresses, and distributed to all delegators in the same way as blocking reward. ### Inavailability The liveness of BSC relies on everyone in the Proof of Staked Authority validator set can produce blocks timely when it is their turn. Validators can miss their turn due to any reason, especially problems in their hardware, software, configuration or network. This instability of the operation will hurt the performance and introduce more indeterministic into the system. There can be an internal smart contract responsible for recording the missed blocking metrics of each validator. Once the metrics are above the predefined threshold, the blocking reward for validator will not be relayed to BC for distribution but shared with other better validators. In such a way, the poorly-operating validator should be gradually voted out of the validator set as their delegators will receive less or none reward. If the metrics remain above another higher level of threshold, the validator will be dropped from the rotation, and this will be propagated back to BC, then a predefined amount of BNB would be slashed from the **self-delegated** BNB of the validator. Both validators and delegators will not receive their staking rewards. ### Governance Parameters There are many system parameters to control the behavior of the BSC, e.g. slash amount, cross-chain transfer fees. All these parameters will be determined by BSC Validator Set together through a proposal-vote process based on their staking. Such the process will be carried on BC, and the new parameter values will be picked up by corresponding system contracts via a cross-chain communication. # Relayers Relayers are responsible to submit Cross-Chain Communication Packages between the two blockchains. Due to the heterogeneous parallel chain structure, two different types of Relayers are created. ## BSC Relayers Relayers for BC to BSC communication referred to as “**BSC Relayers**”, or just simply “Relayers”. Relayer is a standalone process that can be run by anyone, and anywhere, except that Relayers must register themselves onto BSC and deposit a certain refundable amount of BNB. Only relaying requests from the registered Relayers will be accepted by BSC. The package they relay will be verified by the on-chain light client on BSC. The successful relay needs to pass enough verification and costs gas fees on BSC, and thus there should be incentive reward to encourage the community to run Relayers. ### Incentives There are two major communication types: 1. Users triggered Operations, such as `token bind` or `cross chain transfer`. Users must pay additional fee to as relayer reward. The reward will be shared with the relayers who sync the referenced blockchain headers. Besides, the reward won't be paid the relayers' accounts directly. A reward distribution mechanism will be brought in to avoid monopolization. 2. System Synchronization, such as delivering `refund package`(caused by failures of most oracle relayers), special blockchain header synchronization(header contains BC validatorset update), BSC staking package. System reward contract will pay reward to relayers' accounts directly. If some Relayers have faster networks and better hardware, they can monopolize all the package relaying and leave no reward to others. Thus fewer participants will join for relaying, which encourages centralization and harms the efficiency and security of the network. Ideally, due to the decentralization and dynamic re-election of BSC validators, one Relayer can hardly be always the first to relay every message. But in order to avoid the monopolization further, the rewarding economy is also specially designed to minimize such chance: 1. The reward for Relayers will be only distributed in batches, and one batch will cover a number of successful relayed packages. 2. The reward a Relayer can get from a batch distribution is not linearly in proportion to their number of successful relayed packages. Instead, except the first a few relays, the more a Relayer relays during a batch period, the less reward it will collect. ## Oracle Relayers Relayers for BSC to BC communication are using the “Oracle” model, and so-called “**Oracle Relayers**”. Each of the validators must, and only the ones of the validator set, run Oracle Relayers. Each Oracle Relayer watches the blockchain state change. Once it catches Cross-Chain Communication Packages, it will submit to vote for the requests. After Oracle Relayers from ⅔ of the voting power of BC validators vote for the changes, the cross-chain actions will be performed. Oracle Replayers should wait for enough blocks to confirm the finality on BSC before submitting and voting for the cross-chain communication packages onto BC. The cross-chain fees will be distributed to BC validators together with the normal BC blocking rewards. Such oracle type relaying depends on all the validators to support. As all the votes for the cross-chain communication packages are recorded on the blockchain, it is not hard to have a metric system to assess the performance of the Oracle Relayers. The poorest performer may have their rewards clawed back via another Slashing logic introduced in the future. # Outlook It is hard to conclude for Binance Chain, as it has never stopped evolving. The dual-chain strategy is to open the gate for users to take advantage of the fast transferring and trading on one side, and flexible and extendable programming on the other side, but it will be one stop along the development of Binance Chain. Here below are the topics to look into so as to facilitate the community better for more usability and extensibility: 1. Add different digital asset model for different business use cases 2. Enable more data feed, especially DEX market data, to be communicated from Binance DEX to BSC 3. Provide interface and compatibility to integrate with Ethereum, including its further upgrade, and other blockchain 4. Improve client side experience to manage wallets and use blockchain more conveniently ------ [1]: BNB business practitioners may provide other benefits for BNB delegators, as they do now for long term BNB holders.
AKAGIwyf / UAV TrackingIn recent years, UAV began to appear in all aspects of production and life of human society, and has been widely used in aerial photography, monitoring, security, disaster relief and other fields. For example, UAV tracking can be used for urban security, automatic cruise to find suspects and assist in intelligent urban security management.However, the practical application of UAV in various early scenes was mostly based on human remote control or intervention, and the degree of automation was not high. The degree to which UAVs can be automated is one of the decisive factors in whether they can play a bigger role in the future. With the increasing demand of UAV automation, target tracking based on computer vision has become one of the current research hotspots. Some companies in China and abroad, such as DJI, have successfully equipped target tracking on UAVs, but these technologies only exist in papers and descriptions, and the specific implementation has not been sorted out and opened source. Therefore, we plan to try to complete this project by ourselves and open source it on Github. Traditional visual tracking has many advantages, such as strong autonomy, wide measurement range and access to a large amount of environmental information, it also has many disadvantages.It requires a powerful hardware system. In order to obtain accurate navigation information, it needs to be equipped with a high-resolution camera and a powerful processor. From image data acquisition to processing, huge data operations are involved, which undoubtedly increases the cost of UAV tracking. Moreover, the reliability of traditional visual navigation and tracking is poor, and it is difficult for UAV to work in complex lighting and obstacle scenes. Therefore, we plan to use deep learning for target tracking in this project. We can train our own model through deep learning algorithm (we have not decided what network structure to use), then move the trained model to the embedded development board for operation, fix it on the UAV, read the image through the camera and process the data, so that it can recognize the objects to be recognized and tracked. In this project, we will use NVIDIA Jetson TX2 development board, install ROS in Linux system, establish communication with pixhawk, and conduct UAV flight control through PID algorithm.
NaveenKaliannan / StructureFactorA simple matlab code to compute the structure factor S(q) from pair correlation function g(r)
judyboon / BASSA Bayesian group factor model with structured sparsity
tyst3273 / Pynamic Structure Factorpython code to calculate inelastic-neutron-scattering dynamic structure factor, S(Q,w), from molecular dynamics trajectories using parallelism over Q-points.
