SLR-CV

Download Videos(only if needed)

• If you need to download the videos using the MS-ASL dataset, navigate to the data_downloader.ipynb file and run all the cells there. This is a lengthy process. However, if you are cloning the code from GitHub, you should already have access to all the videos and you will not need to use this file at all
  • Create a new Python environment, install the dependencies in requirements.txt using pip install -r requirements.txt.
  • Inside this file, you can specify the save folder in the download_videos function by changing SAVE_PATH = <your folder name>.
  • You can choose the input to the download_videos function from either test_data100, train_data100, or val_data100.

Demo Code

• For a dummy run, navigate to dummy_demo.ipynb file. Run all the cells.
• In this file, we selected only 5 classes and augmented each image 3 times to shorten the code running process.
• The max training epoch is also set to 5 for the demo.
• This demo is only for 3-Channel Dataset as 5 Channel dataset takes a little time to load. The process is all the same.
• This process should take about less than 5 minutes if trained on GPU.

For the project code, check train_3Channel.ipynb and train_5Channel.ipynb files

This github repo is about 1.2GB.