A set of neural network models and visualizations developed for the practical course on Neural Networks at the Karlsruhe Institute of Technology.
git clone --recurse-submodules git@github.com:benjaminalt/signlanguage-recognition.git
We recommend to use the Python distribution Anaconda and a custom virtual environment:
conda create --name signlanguage-env
source activate signlanguage-env
conda install graphviz python-graphviz matplotlib ipython opencv pandas scipy
conda install pytorch torchvision -c pytorch
cd signlanguage-recognition
pip install -r requirements.txt
Before running anything from the repository, make sure the right virtual environment is active:
source activate signlanguage-env
To train the neural network, simply run main.py:
cd signlanguage-recognition/code
python main.py train
This will create a timestamped directory in signlanguage-recognition/results, which contains
- A CSV file with the parameters of the neural network and the training results
- A visual representation of the neural network
- A matplotlib plot of training and test loss and accuracy
- A pytorch (.pt) file containing the weights of the model
You can tweak the parameters of the model by adapting code/options.py.
cd signlanguage-recognition/code
python main.py test --weights path/to/weights.pt --data path/to/data.csv
This will construct a model with the given weights and use it to classify the provided data. The resulting accuracy will be printed.
cd signlanguage-recognition/code
python main.py visualize --weights path/to/weights.pt --data path/to/data.csv
This will generate a Grad-CAM visualization for all convolutional layers of the model for all input images. A lot of images will be generated, so choose your input data wisely (use for example data/sign_mnist_test_small.csv). The results will be in a generated folder under results.