15 skills found
zjhellofss / KuiperInfer校招、秋招、春招、实习好项目！带你从零实现一个高性能的深度学习推理库，支持大模型 llama2 、Unet、Yolov5、Resnet等模型的推理。Implement a high-performance deep learning inference library step by step
oborchers / Fast Sentence EmbeddingsCompute Sentence Embeddings Fast!
evarae / CNN TutorialHi! Thanks for checking out my tutorial where I walk you through the process of coding a convolutional neural network in java from scratch. After building a network for a university assignment, I decided to create a tutorial to (hopefully) help others do the same and improve my own understanding of neural networks.
jiajingchen113322 / Recycle Maxpooling ModuleThis is the implementation of Recycle Maxpooling Module for Point Cloud Analysis
subham2203 / Reimagined WinnerCIFAR-10 Object Detection with improved accuracy using Fractional MaxPooling with Convolutional Neural Networks
Tirth8038 / Multiclass Image Classification The main aim of the project is to scan the X-rays of human lungs and classify them into 3 given categories like healthy patients, patients with pre-existing conditions, and serious patients who need immediate attention using Convolutional Neural Network. The provided dataset of Grayscale Human Lungs X-ray is in the form of a numpy array and has dimensions of (13260, 64, 64, 1). Similarly, the corresponding labels of X-ray images are of size (13260, 2) with classes (0) if the patient is healthy, (1) if patient has pre-existing conditions or (2) if patient has Effusion/Mass in the lungs. During data exploration, I found that the class labels are highly imbalanced. Thus, for handling such imbalanced class labels, I used Data augmentation techniques such as horizontal & vertical flips, rotation, altering brightness and height & width shift to increase the number of training images to prevent overfitting problem. After preprocessing the data, the dimension of the dataset is (31574, 64, 64, 1). For Model Selection, I built 4 architectures of CNN Model similar to the architecture of LeNet-5, VGGNet, AlexNet with various Conv2D layers followed by MaxPooling2D layers and fitted them with different epochs, batch size and different optimizer learning rate. Moreover, I also built a custom architecture with comparatively less complex structure than previous models. Further to avoid Overfitting, I also tried regularizing Kernel layer and Dense layer using Absolute Weight Regularizer(L1) and to restrict the bias in classification, I used Bias Regularizer in the Dense layer. In addition to this, I also tried applying Dropout with a 20% dropout rate during training and Early Stopping method for preventing overfitting and evaluated that Early Stopping gave better results than Dropout. For evaluation of models, I split the dataset into training,testing and validation split with (60,20,20) ratio and calculated Macro F1 Score , AUC Score on test data and using the Confusion Matrix, I calculated the accuracy by dividing the sum of diagonal elements by sum of all elements. In addition to this, I plotted training vs. validation loss and accuracy graphs to visualize the performance of models. Interestingly, the CNN model similar to VGGNet with 5 Conv2D and 3 MaxPooling layers and 2 Dense layers performed better than other architecture with Macro F1 score of 0.773 , AUC score of 0.911 and accuracy of 0.777.
Abhi-H / CNN With Fashion MNIST DatasetThis is my first implementation of a network. I have used a 2 layered CNN with maxpooling and Xavier weights.
LLeon360 / UNetBiomedicalDiagnosisContains a module for functions that build and evaluate a U-Net Model which is a modified auto-encoder image-to-image architecture that includes skip connections from encoder to decoder layers with matching input dimensions to retain features from the encoded image. The model is designed for image segmentation and produces a mask based off x-ray data. The model is used with lung x-rays to segment out lungs, brain scans to segment out brain tumors, and pictures of a room to segment out a human figure. With the lungs dataset, it is able to manage 98% accuracy with a wide range of 3,4,5 encoding block of Conv2D and MaxPooling and BatchNormalization + 1 latent layer without pooling and the mirrored number of decoding layers that do Conv2DTranspose for deconvolution and UpScaling to reverse MaxPooling making 7,9,11 layers with doubling filters/kernels in each encoding block from 16,32,64,80 that double with each block like in a traditional CNN and then half with each decoding block.
sushant1827 / Fashion Clothing ClassificationKaggle Machine Learning Competition Project : In this project, we will create a classifier to classify fashion clothing into 10 categories learned from Fashion MNIST dataset of Zalando's article images
yzldw333 / Neural NetworkThis a python neural network. (Input Layer, Convolution Layer, MaxPooling Layer ,AvgPoolingLayer, Full-Connected Layer,Sigmoid Layer, TanhLayer, Output Layer(LMS and SoftMax))
tejaslodaya / Tensorflow Signs CnnAn algorithm that facilitates communication between a speech-impaired person and someone who doesn't understand sign language using convolution neural networks
amirrezarajabi / Neural Network Implementation From Scratchimplementation of neural network from scratch only using numpy (Conv, Fc, Maxpool, optimizers and activation functions)
AlwaysDhruv / Image Classification CPPHi their my self Dhruv. So this repository or project are developed on C++ and Python for image recognize. C++ are main engine and python are work preprocessing only. more information are in README file.
devsonni / Chart Classification Using CNN KerasThis repository consists of models of CNN for classifying different types of charts. Moreover, it also includes script of fine-tuned VGG16 for this task. On top of that CradCAM implementation of fine-tuned VGG16.
MostafaFiroozi / HAND GESTURE CLASSIFICATION BASED ON SURFACE EMG SIGNALS FOR ONLINE HAND PROSTHETICS CONTROLWe implemented 5 classical classifiers and a CNN to classify 8 different hand gestures. The classical classifiers we chose are the one in the picture and for each of them we extracted 4 time domain features, that showed to be effective in literature. This is the architecture of the CNN we used: in input we have one window of the raw EMG signal, with all the 10 channels. Then there are three main blocks, each composed by one convolutional layer and one maxpooling layer. At the output we have the predicted class of the gesture.To make the results more consistent, we analyzed 3 different day sessions of the same subject: for each of them we explored 7 different window sizes (ranging from 25 to 300ms) and 4 different stride lengths (from 25 to 100% of the window size). The performances of the classifiers are assessed in terms of the trade off between accuracy (that should be as high as possible) and computational time (which must be lower than the real time constraint of 300ms). This graph shows the results of our CNN: exploiting window sizes higher than 100ms, does not significantly affect the accuracy, whereas smaller window sizes lead to a decrease of accuracy. Moreover, for each window size, the use of lower stride lengths enhanced the accuracy (indeed it is a data augmentation tool). In addition, the computational time grows with the increase of the window size and of the stride length.SVM results, which is the best classifier among the 5 we used. The trend of accuracy and computational time is very similar to the CNN one, even though, in this case the stride length doesn’t affect the computational time. In conclusion, we present the trend of mean accuracy for the classical classifiers compared to CNN. In general, we can notice a saturation behavior, which for CNN starts from around 100ms. The computational time of all classifiers shows a linear increasing trend, despite the CNN being characterized by a lower order of magnitude. Knowing that for an online application, the computational time must also include the window size, in our opinion the best trade off for the CNN between accuracy and computational time is reached at 100ms.
