This repository was forked from Dongxu Li et al.'s WLASL repository, whose dataset was made open source for research purposes. This project investigates the I3D model's usability for a mobile application. A demo of the mobile app we've developed so far, called Sign-a-mander, can be found here.
When you clone this repository, you will need to download a few zip files for the I3D model to work.
archived - Put this folder inside the I3D folder. It has the weights for asl10 to asl2000.
weights - Also put this folder inside the I3D folder. These are the "I3D weights pre-trained Kinetics" in the original README file. It's referenced in the def ensemble
function in the predict, test, and train I3D files, but according to README-WLASL.md, it is used for training.
Note: I downloaded Li et al.'s entire dataset via Kaggle: https://www.kaggle.com/datasets/risangbaskoro/wlasl-processed.
Please refer to the original repository's README for additional instructions. You can also find it here with the file name README-WLASL.md
.