Classifier for images of letters in American Sign Language. Written to help me learn CNNs and PyTorch.

Dependencies:

• python
• make
• rsync
• p7zip
• kaggle-api

Make sure the credentials for kaggle-api are configured.

Install python dependencies:

$ pip -r requirements.txt

Download the data:

$ make

Then train the model:

$ ./model.py train

This will save a *.pt model file every five epochs in build/<training start time>/. For example, the directory might be called 2020-12-03T12:40:13.

Run the tests on a chosen model file:

$ ./model.py test build/<start time>/<file name>.pt

The server will load the file model.pt in the project root. So just copy (or symlink) the chosen model.

$ cp build/<start time>/<file name>.pt ./model.pt

Run the web server:

$ flask run

Finally, open http://localhost:5000 in your browser.