Cepheid decouples scale from structure for cross-family sparse-to-dense motion completion
NYCU · hf · 2026-10-14
NYCU researchers present Cepheid for sparse-to-dense mesh motion completion:
- Residual priors trained on one geometry family often fail off-family vs parameter-free harmonic extension; cross-family training hits a scaling barrier as residual amplitude varies over an order of magnitude.
- The fix: learn a dimensionless prior decoupling residual structure from physical scale, estimating amplitude from observables plus a lightweight correction fitted only on training sources.
The unified model beats harmonic extension across all evaluated families, including unseen cloth and legged robots, without skeletons or correspondence, and transfers zero-shot to point-tracked animal video and soft-body robot grasping.
More from Embodied
- Nature: smart glasses are heading for a major privacy crisis — marigo · 2026-10-14
- Impatient for JetKVM Mini, dev buys LuckFox KVM to let agents control his Mac Mini — TheZachMueller · 2026-10-14
- China's electric autonomous tractor T70 goes viral, already deployed in Hebei — lukas_m_ziegler · 2026-10-14
- Nomagic teases more updates as warehouse robots tackle unexpected scenarios — m_wulfmeier · 2026-10-14
- NTU team has Claude write robot control code, hitting 70.24% success in sim — aigclink · 2026-10-14
- Rabbit R1 called 'ahead of its time' as modders plan to hack the once-mocked AI device — SimonBalmain · 2026-10-14