MambaGait: Gait recognition approach combining explicit representation and implicit state space model

Open Webpage

Haijun Xiong, Bin Feng, Bang Wang, Xinggang Wang, Wenyu Liu

Published: 2025, Last Modified: 20 Feb 2026Image Vis. Comput. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•First approach to explore the potential of Mamba in gait recognition, with state-of-the-art performance on four challenging datasets (GREW, Gait3D, CCPG, and SUSTech1K).•ESF (Explicit Spatial Representation Field) enhances the representation of binary silhouettes and endows them with spatial position awareness, thereby improving the sensitivity of most regions to spatiotemporal dynamics and providing richer information representation.•TWSM (Temporal Window Switch Mamba Block) leverages bidirectional temporal window switching to establish cross-window communication, implicitly capturing long-range dependencies while extracting representative local spatiotemporal features.