Reproduce “Human Mesh Modeling for Anny Body” 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

Romain Brégier, Guénolé Fiche, Laura Bravo-Sánchez, Thomas Lucas, Matthieu Armando, Philippe Weinzaepfel, Grégory Rogez, Fabien Baradel

Anny is presented as a novel parametric human body model that captures diverse human shapes and ages, built using open-source assets and calibrated with WHO population statistics. The model demonstrates competitive accuracy in 3D scan fitting (2.5mm error for adults, 2.0mm for children) and, when integrated with a synthetic dataset (Anny-One), enhances Human Mesh Recovery performance, particularly for child body shapes on the AGORA-Kids benchmark (PVE 113.9).

Key insights

  • Parametric human body models can be effectively constructed from open-source artistic priors and anthropometric statistics, eliminating the need for expensive, proprietary, and privacy-sensitive 3D scan datasets.
  • Grounding shape spaces in interpretable phenotype parameters allows for precise semantic control over human morphology and enables the creation of demographically representative models spanning the entire human lifespan.
  • High-quality synthetic datasets generated from anthropometrically sound and interpretable models can match or exceed the performance of real scan-based data for training Human Mesh Recovery (HMR) models, especially in scenarios requiring broad demographic diversity.

Summary by alphaXiv.

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