Reproduce “PLACE: Proximity Learning of Articulation and Contact in 3D Environments” 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
Siwei Zhang, Yan Zhang, Qianli Ma, Michael J. Black, Siyu Tang
Researchers from ETH Zürich and the Max Planck Institute for Intelligent Systems developed PLACE, a method for synthesizing realistic 3D human bodies in 3D environments by explicitly modeling human-scene proximity and contact using a geometry-based representation. This approach yielded state-of-the-art perceptual naturalness in user studies, with generated interactions rated as highly plausible and often indistinguishable from real human-scene interactions, alongside improved physical plausibility and diversity.
Key insights
- Explicitly modeling 3D proximity and contact through a consistent, geometry-based BPS representation enables the synthesis of more realistic and physically plausible human-scene interactions without semantic labels.
- A cVAE architecture can effectively learn to generate diverse and natural human body poses that cohere with a given 3D scene by conditioning on the proposed proximity representation.
- Post-processing optimization, which leverages collision and contact terms, is crucial for significantly enhancing the physical plausibility, diversity, and overall naturalness of the generated human-scene interactions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.