Combo-Gait: Unified Transformer for Multi-Modal Gait Recognition and Attribute Analysis
A new framework, Combo-Gait, combines 2D silhouettes and 3D SMPL features for robust gait recognition. It employs a unified transformer to fuse multi-modal data and jointly performs identification with attribute estimation (age, BMI, gender). The approach addresses limitations of single-modality methods in capturing full walking dynamics.
Key facts
- Combo-Gait is a multi-modal and multi-task framework for gait analysis.
- It combines 2D temporal silhouettes with 3D SMPL features.
- A unified transformer fuses multi-modal gait features.
- Jointly performs gait recognition and human attribute estimation.
- Attributes include age, body mass index (BMI), and gender.
- Aims to improve robustness under low-resolution or unconstrained environments.
- Single-modality methods often fail to capture full geometric and dynamic complexity.
- Paper published on arXiv with ID 2510.10417.
Entities
Institutions
- arXiv