Uni4R: Unifying Continuous-Time 4D Reconstruction and Tracking via Optimal Transport and ODEs
zhenjun_zhao · x · 2026-08-13
Existing 4D reconstruction and point tracking approaches rely on heuristic interpolations or predict only at integer timestamps, lacking kinematic coherence. This paper proposes Uni4R, a framework unifying these tasks by learning continuous velocity fields through the synergy of Optimal Transport (OT) and Ordinary Differential Equodes (ODE).
- Core Mechanism: The continuous velocity field acts as a kinematic prior that mutually benefits both 4D reconstruction and point tracking.
- FMGD: Introduces the Flow Matching Guided Decoder. A global velocity branch extracts anchor features capturing global dynamics, while Flow Matching theory formulates a probability path defined by OT on the anchor feature manifold for velocity prediction.
- Training: Proposes an interpolation strategy to overcome the absence of ground-truth velocities in fractional frames during training.
Related event: Uni4R Unifies 4D Reconstruction and Tracking via OT and ODE(2 posts)→
More from Research
- New RLVR Method Uses Parameter-Space Exploration to Stabilize LLM Training — BayesRL · 2026-08-13
- Manifold launches to accelerate robotics research evaluation, compressing a month of experiments into a week — ZeYanjie · 2026-08-13
- DeepMind Paper Maps 4 Technical Pathways from AGI to ASI — rohanpaul_ai · 2026-08-13
- Researchers' predicted milestones for automated AI research already falling — The Decoder · 2026-08-13
- Paper2Agent: Open-Source System Turns Research Papers into Interactive AI Agents — tom_doerr · 2026-08-13
- Selective imperfection: symmetry breaking as recursive generative mechanism across biology, materials, and music — ProfBuehlerMIT · 2026-08-13