Reproduce “BEST: BERT Pre-Training for Sign Language Recognition with Coupling Tokenization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Weichao Zhao, Hezhen Hu, Wengang Zhou, Jiaxin Shi, Houqiang Li
BEST does isolated sign-language recognition by predicting glosses from short signs. It discretizes upper-body and hand poses into pseudo-labels, then trains a Transformer to predict masked parts from sequence context. On WLASL2000, this pretraining reaches 46.25% Top-1 per-instance accuracy, versus 27.13% without pretraining; the result is benchmark-specific and relies on estimated 2D poses.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.