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

TL;DR

HaGRIDv2 is an expanded and improved hand gesture dataset with 1 million images, enhancing dynamic gesture recognition, reducing false positives, and improving model generalization for gesture and hand detection tasks.

Contribution

The paper introduces HaGRIDv2, a comprehensive hand gesture dataset with new gestures and improved recognition algorithms, surpassing previous datasets in accuracy and generalization.

Findings

Outperforms the original HaGRID in pre-training tasks.

Reduces false positives by 6 times.

Achieves the best generalization among gesture datasets.

Abstract

This paper proposes the second version of the widespread Hand Gesture Recognition dataset HaGRID -- HaGRIDv2. We cover 15 new gestures with conversation and control functions, including two-handed ones. 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 diversified by adding samples of natural hand movements, which allowed us to minimize false positives by 6 times. Combining extra samples with HaGRID, the received version outperforms the original in pre-training models for gesture-related tasks. Besides, we achieved the best generalization ability among gesture and hand detection datasets. In addition, the second version enhances the quality of the gestures generated by the diffusion model. HaGRIDv2, pre-trained models, and a dynamic gesture recognition algorithm are publicly available.