Reproduce “Score-Guided Diffusion for 3D Human 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

Anastasis Stathopoulos, Ligong Han, Dimitris Metaxas

ScoreHMR introduces a method for 3D human pose and shape estimation that leverages diffusion models to refine initial regression estimates. The approach treats recovery as a score-guided diffusion process, improving image-model alignment and temporal consistency across single-frame, multi-view, and video applications.

Key insights

  • Diffusion models can effectively serve as learned priors for complex inverse problems like 3D human pose and shape estimation.
  • Combining initial regression estimates with score-guided diffusion allows for efficient and accurate refinement, bridging the gap between regression and optimization.
  • A task-agnostic diffusion model, once trained, can be applied to diverse 3D human recovery scenarios (single-frame, multi-view, video) by adjusting guidance terms.

Summary by alphaXiv.

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