Reproduce “UT-HMR: Uncertainty-Guided Temporal Modeling for Robust 3D Human Mesh Recovery” 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
Zhibo Wang, Sirisha Rambhatla, Yuhao Chen, Maria Koshkina, David Clausi
UT-HMR uses adjacent frames and uncertainty-aware training to improve multi-person 3D mesh recovery from full images. It fuses three frames before pose decoding and weights better-constrained joints more in an auxiliary loss, while retaining uniform supervision. On VIP-HockeyPose-v2, errors fall versus single-frame SAT-HMR, but inference takes 39.5 rather than 28.1 milliseconds per frame; this comparison does not isolate each component.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.