Cepheid: Decoupling Scale from Structure for Cross-Family Motion Completion
Cepheid targets a cross-family failure mode in sparse-to-dense motion completion: residual priors that lose to harmonic extension off-domain.
The paper says residual amplitude shifts by more than an order of magnitude across geometry families and shrinks as observations get denser. Its method separates residual structure from physical scale, then estimates amplitude from observable signals with a lightweight correction trained on source families. The authors report one unified model beating harmonic extension across evaluated families, including unseen cloth and legged bodies, without skeletons, registration, or cross-mesh correspondence. They also apply the same prior without retraining to point-tracked animal video and a robot soft-body grasping case. Source: HF Daily Papers' note.
The paper says residual amplitude shifts by more than an order of magnitude across geometry families and shrinks as observations get denser. Its method separates residual structure from physical scale, then estimates amplitude from observable signals with a lightweight correction trained on source families. The authors report one unified model beating harmonic extension across evaluated families, including unseen cloth and legged bodies, without skeletons, registration, or cross-mesh correspondence. They also apply the same prior without retraining to point-tracked animal video and a robot soft-body grasping case. Source: HF Daily Papers' note.
score 3