Reproduce “SCR-Mamba: Skeleton-conditioned Mamba with cross-scale incremental refinement for efficient 3D human pose estimation” 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

Jie Wang, Xuejing Gu, Congzhe Zhang, Zibo Zhao

SCR-Mamba estimates 3D human poses from 2D joint tracks in monocular video. It uses bone and left–right symmetry context to guide spatial and temporal state updates, then refines retained frame features with changes from shorter temporal scales. On 243-frame Human3.6M clips, SCR-Mamba-L reports 36.8 mm P1 error using 15.10 G MACs, versus PoseMamba-L’s 38.1 mm and 47.93 G; MACs measure computation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.