Reproduce “MMGait: Towards Multi-Modal Gait Recognition” 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
Chenye Wang, Qingyuan Cai, Saihui Hou, Aoqi Li, Yongzhen Huang
A new comprehensive multi-modal gait benchmark, MMGait, was introduced, integrating data from five heterogeneous sensors and twelve derived modalities to address limitations of existing datasets. The OmniGait framework, capable of unified multi-modal recognition, demonstrated competitive performance across single-modal tasks (e.g., RGB NM R1: 99.7%) and achieved strong cross-modal (e.g., RGB→Depth R1: 87.0%) and multi-modal fusion results.
Key insights
- Different sensing modalities offer complementary information, with IR and structural modalities (Depth, LiDAR) demonstrating increased robustness to clothing variations compared to RGB-based approaches.
- A single unified model can effectively learn a shared embedding space across diverse heterogeneous modalities, performing competitively across single-modal, cross-modal, and multi-modal tasks.
- The proposed Omni Multi-Modal Gait Recognition task is feasible and addresses a practical need for flexible and efficient gait recognition systems adaptable to various sensor inputs and retrieval scenarios.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.