For detailed explanation, please check out our paper: 👐OpenHands: Making Sign Language Recognition Accessible with Pose-based Pretrained Models across Languages

Currently supported

  • Easily train pose-based isolated sign language recognizers (ISLR)
    • Currently supports 4 different models

  • Support for many run-time augmentations for pose keypoints
    • Check for complete list of pose transforms

  • Support for efficient inference
    • Check for more details:

  • Easily pre-train models using monolingual SL data
    • And fine-tune on any small ISLR dataset

  • All Training and Inference is completely config-based
    • No-code required

    • Check for example configs:

  • Supports 6 sign languages out-of-the-box
    • Check for supported datasets:

Future Plan

  • [ ] Add more sign langauges

  • [ ] Explore more SoTA models

  • [ ] Support for sign language detection

  • [ ] Support for continuous sign langauge recognition (CSLR)