ASL

The SignBank Project

Authors:

• Thomas Fouts (University of Michigan)
• Ali Hindy (Stanford)
• Chris Tanner (Harvard)

To Run Models:

python3 ASL.py

The program accepts a few hyperparameters:

--model_type (one of {LSTM, BILSTM, CNN, BILSTMCNN})

--epochs (# of epochs)

--hidden_size (# of units in the 1st hidden layer)

--ensemble (True or False)

Our best performing model yielded 95.9 accuracy and can be run via:

python3 ASL.py -m BILISTMCNN -e 75 -hs 768 -e True

Repository Structure

• All data is in the "data" folder. To generate new data, run the Arduino Script "DataCollector.ino"
• All models are in the "src" folder.
• To view the schematics in order to build one of our sensors yourself, view the "schematics" folder.