Code Monkey home page Code Monkey logo

paulstreli's Projects

3d-hand-pose-sequence-data-augmentation-using-gans icon 3d-hand-pose-sequence-data-augmentation-using-gans

The human hand plays a crucial role in conveying emotions and carrying out most day-to-day activities. Therefore numerous modern technologies - ranging from gesture control to autonomous driving - would benefit from the reliable recognition of certain hand actions. This can be done using a two-step approach, in which first hand poses are obtained from video frames and then the resulting sequences are classified in the 3D skeleton space. Existing techniques that aim to solve the second step are mostly based on deep learning methods. Given the high complexity and dimensionality of the human hand, these require large amounts of training data to achieve good performance. As the collection of precisely annotated hand pose data is time-consuming and expensive, data augmentation appears as an advantageous practice to increase the recognition accuracy for a given classifier. This thesis proposes a suitable WGAN-GP architecture for the generation of synthetic hand skeleton sequences with variable length. The recommended critic consists of a multi-layer perceptron with three hidden layers, while the generator is based on two RNNs and receives a start frame as input. Both networks are conditioned on the action class. The best performing model was trained on multiple classes simultaneously and selected based on the smallest generator loss. When its synthetic samples were used to augment the training set of a 1-layer LSTM classifier, the classification error on several subsets as well as on the complete dataset decreased. Quantitative results show that the chosen GAN-based data augmentation outperforms alternative standard methods. Furthermore, no clear correlation between the visual appearance of the generated samples and their resulting improvement on recognition accuracy was found.

distance-metric-learning---cuhk03-dataset icon distance-metric-learning---cuhk03-dataset

This repository contains the results for my second Pattern Recognition Coursework that I completed with my partner Karoly. The report describes person re-ID experiments on the popular CUHK03 dataset that contains pictures of pedestrians taken from two different surveillance cameras. A set of features was already provided. The goal is to find a suitable distance metric which learns a feature transformation yielding improved performance on a range of metrics for kNN-retrieval. The original covariance based Mahalanobis method, Large Margin Nearest Neighbour Distance Metric, Metric Learning for Kernel Regression and a fully connected Neural Network with Triplet Loss are considered for this purpose. The Kernel Regression gave the best results with a performance similar to the untransformed baseline approach.

face-recognition-pca-lda icon face-recognition-pca-lda

This repository contains the final results for my first Pattern Recognition Coursework that I completed with my partner Karoly. Face recognition is the problem of identifying a human, based on a picture of their face. In this coursework various face-recognition algorithms are investigated, including Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) and PCA-LDA Ensemble in different variations. The recognition accuracies of the models are compared and advantages and disadvantages are determined.

mnist-gans icon mnist-gans

This repository contains the final results for my second Computer Vision Coursework that I completed with my partner Karoly.

pytorch icon pytorch

Tensors and Dynamic neural networks in Python with strong GPU acceleration

wine-quality-prediction icon wine-quality-prediction

As part of my third-year module Machine Learning I wrote a four-page paper with the name "Analysis of Regression Predictors for the Wine Quality Dataset". The paper presented my results on finding a suitable hypothesis that is able to predict a quality score (between 0 and 10) based on eleven physicochemical attributes of a specific wine. For this regression problem I implemented several machine learning techniques and algorithms using Matlab and experimented with constant base predictors, support vector machine regression, neural networks, bag of trees and linear regression with and without feature transform and regularisation. For the total dataset of 6497 samples provided by the UCI Machine Learning Repository bag of trees seems to yield the best performance.

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.