HaGRIDv2: 1M Images for Static and Dynamic Hand Gesture Recognition

This paper proposes the second version of the widespread image-based Hand Gesture Recognition dataset HaGRID -HaGRIDv2.We have added 15 new gestures to the existing 18, encompassing both conversational and control functions, including two-handed gestures.Building on the foundational concepts proposed by HaGRID's authors, we implemented the dynamic gesture recognition algorithm and further enhanced it by adding three new groups of manipulation gestures.The "no gesture" class was significantly expanded and redefined with a new semantic focus to include a diverse range of natural hand movements, which led to a 16-fold reduction in false positives on HaGRIDv2.The HaGRIDv2 dataset outperforms the original HaGRID in pre-training models for gesturerelated tasks.Besides, we achieved the best generalization ability among gesture and hand detection datasets.Additionally, the second version of the dataset provides a diverse range of hand samples, which is crucial for fine-tuning modern diffusion models.By fine-tuning on HaGRIDv2, these models achieve improved outcomes in generating anatomically correct hand gesture images.HaGRIDv2, pre-trained models, and a dynamic gesture recognition algorithm are publicly available.