UniMate: One Unified Model to Animate Diverse Skeletons
UniMate claims one text-driven animation model can handle arbitrary rigged skeletons without per-skeleton tuning.
The paper introduces a topology-aware diffusion transformer that uses skeletal graph relations, Laplacian-based position encoding, and a rest-pose topology conditioner. The authors also built UniML3D, a 13,006-sequence dataset spanning bipeds, quadrupeds, birds, marine forms, insects, serpentine bodies, and articulated rigid objects. Trained on that set, UniMate is reported to beat state-of-the-art baselines on quality, generalization, and efficiency, with support for zero-shot cross-topology transfer, in-betweening, expansion, and text-guided editing. ArXiv · AI/CL/LG's note
The paper introduces a topology-aware diffusion transformer that uses skeletal graph relations, Laplacian-based position encoding, and a rest-pose topology conditioner. The authors also built UniML3D, a 13,006-sequence dataset spanning bipeds, quadrupeds, birds, marine forms, insects, serpentine bodies, and articulated rigid objects. Trained on that set, UniMate is reported to beat state-of-the-art baselines on quality, generalization, and efficiency, with support for zero-shot cross-topology transfer, in-betweening, expansion, and text-guided editing. ArXiv · AI/CL/LG's note
score 5