Reproduce “GeneOH Diffusion: Towards Generalizable Hand-Object Interaction Denoising via Denoising Diffusion” 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
Xueyi Liu, Li Yi
Researchers from Tsinghua University and Shanghai AI Laboratory developed GeneOH Diffusion, a method for denoising hand-object interaction data by introducing a generalized contact-centric representation (GeneOH) and a two-step diffusion-based denoising strategy. The method consistently outperformed baselines on various benchmarks and demonstrated strong generalization to unseen noise patterns and datasets, providing cleaner data for applications like video-based hand tracking and motion retargeting.
Key insights
- Canonicalization of hand-object interactions into a generalized contact-centric representation (GeneOH) promotes generalization across diverse objects and motions by removing global pose dependencies.
- Adapting diffusion models for inverse problems like denoising, by training a 'canonical denoiser' on a whitened noise space, enables remarkable robustness to unseen noise patterns.
- A progressive, multi-stage denoising strategy (MotionDiff, SpatialDiff, TemporalDiff) effectively addresses the heterogeneous and complex nature of interaction noise in hand-object sequences.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.