Reproduce “3D Human Pose, Shape and Texture from Low-Resolution Images and Videos” 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

Xiangyu Xu, Hao Chen, Francesc Moreno-Noguer, Laszlo A. Jeni, Fernando De la Torre

RSC-Net estimates 3D human pose from person crops across resolutions using shared features and resolution-specific fusion weights. It trains predictions at lower resolutions to match higher-resolution outputs and uses contrastive learning to align features across scales. On 32×32 3DPW inputs, it reports MPJPE of 117.12 versus 137.61 for SPIN, but tests use downsampled crops, not raw surveillance or broadcast feeds.

Summary by alphaXiv.

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