This paper addresses the challenging problem of denoising hand-object interactions (HOI) to produce perceptually realistic sequences. The authors propose GeneOH Diffusion, a novel approach that includes two key components: an innovative contact-centric HOI representation named GeneOH and a domain-generalizable denoising scheme. GeneOH encodes the HOI process by incorporating hand trajectory, hand-object spatial relations, and temporal relations, while the denoising scheme consists of a canonical denoiser trained to project noisy data from a whitened noise space to a clean data manifold. The method is evaluated on four benchmarks with significant domain variations, demonstrating superior effectiveness and generalizability. GeneOH Diffusion shows promise for various downstream applications, such as refining noisy hand trajectories and cleaning synthesized motions. The paper also includes extensive experiments, ablation studies, and a detailed discussion of the limitations and future directions.This paper addresses the challenging problem of denoising hand-object interactions (HOI) to produce perceptually realistic sequences. The authors propose GeneOH Diffusion, a novel approach that includes two key components: an innovative contact-centric HOI representation named GeneOH and a domain-generalizable denoising scheme. GeneOH encodes the HOI process by incorporating hand trajectory, hand-object spatial relations, and temporal relations, while the denoising scheme consists of a canonical denoiser trained to project noisy data from a whitened noise space to a clean data manifold. The method is evaluated on four benchmarks with significant domain variations, demonstrating superior effectiveness and generalizability. GeneOH Diffusion shows promise for various downstream applications, such as refining noisy hand trajectories and cleaning synthesized motions. The paper also includes extensive experiments, ablation studies, and a detailed discussion of the limitations and future directions.