Reproduce “TexturePose: Supervising Human Mesh Estimation with Texture Consistency” 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
Georgios Pavlakos, Nikos Kolotouros, Kostas Daniilidis
TexturePose uses clothing consistency across images of the same person as an additional training signal for estimating 3D body pose and shape. It compares colors at corresponding visible body-surface locations, alongside 2D keypoints and an adversarial prior. On Human3.6M Protocol 1, this reduces mean reconstruction error from 93.0 to 80.2 millimeters, but remains above the 64.8-millimeter fully 3D-supervised result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.