Transformer Video Sign-Language Recognition
Transformer-based continuous sign-language recognition and translation from video streams.
Research Intern — with Prof. Xiaofang Liu · Apr 2023 – Jun 2024
- Developed a Transformer-based video-stream sign-language recognition algorithm modeling point similarity and gesture motion across consecutive frames.
- Integrated S2T and S2G+G2T architectures with CTC loss to model word–sentence correlations and improve real-time accuracy.
- Achieved ~87% BLEU-1/BLEU-2; received national recognition in the College Students Innovation and Entrepreneurship Competition.