File size: 1,155 Bytes
2d9fc93 c67d0a1 e35b538 2d9fc93 e35b538 2d9fc93 e35b538 2d9fc93 e35b538 2d9fc93 e35b538 2d9fc93 e35b538 2d9fc93 e35b538 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 | ---
title: SignLanguageTranslate
emoji: 👁
colorFrom: red
colorTo: gray
sdk: docker
app_port: 7860
pinned: false
license: mit
---
# SignLanguageTranslate
SignLanguageTranslate aims to support sign languages around the world. This demo currently evaluates Indian Sign Language on isolated clips from INCLUDE.
The Transformer comparison uses archived predictions from the rolling Transformer at epoch 31 and the causal streaming Transformer at epoch 102. The accuracy table uses 720 isolated INCLUDE clips from the saved iPad validation split. Both models use the same clips. The original LSTM demo uses its historical test manifest.
Public source videos are matched by category, class and filename to a pinned revision of the [public INCLUDE mirror](https://huggingface.co/datasets/spsarolkar/AI4Bharat-INCLUDE-dataset). The app labels that repository's split separately. The new comparison credits the original [AI4Bharat INCLUDE dataset](https://huggingface.co/datasets/ai4bharat/INCLUDE). The [keypoint dataset card](https://huggingface.co/datasets/spsarolkar/AI4Bharat-INCLUDE-keypoints) describes the Apple Vision input used for the comparison.
|