Reproduce “CG-HOI: Contact-Guided 3D Human-Object Interaction Generation” 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
Christian Diller, Angela Dai
CG-HOI introduces the first method to generate dynamic 3D human-object interaction sequences from text descriptions, where both human and object motions are interdependent. The approach leverages explicit contact modeling and guidance during a denoising diffusion process, producing physically plausible interactions with reduced artifacts like object penetration or floating.
Key insights
- Explicitly modeling and guiding via human-object contact distances is crucial for achieving physically plausible and coherent dynamic human-object interactions.
- Jointly modeling human pose (SMPL), object transformation, and their mutual contact within a denoising diffusion framework enables learning complex interdependencies.
- Aggregating multiple object motion hypotheses weighted by the inverse of predicted contact distances allows the object's movement to be driven by the most relevant human body parts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.