SignBart is an efficient Isolated Sign Language Recognition model that decouples x and y coordinates using a lightweight encoder-decoder architecture. It achieves high accuracy with fewer than 750K parameters, outperforming traditional models on datasets like LSA-64, WLASL, and ASL-Citizen.
deep-learning attention-mechanism sign-language-recognition isolated-sign-language-recognition skeleton-based-models bart-encoder-decoder skeleton-keypoints
- Updated
Jun 30, 2025 - Python