Uni4R Marries Optimal Transport and ODEs for Unified 4D Reconstruction
zhenjun_zhao · x · 2026-08-13
The paper proposes Uni4R, a framework that unifies 4D reconstruction and point tracking by learning continuous velocity fields through the synergy of Optimal Transport (OT) and Ordinary Differential Equations (ODE).
Existing methods typically rely on heuristic interpolations, lacking kinematic coherence. The continuous velocity field proposed in Uni4R acts as a kinematic prior that mutually benefits both tasks. At its core, the Flow Matching Guided Decoder (FMGD) extracts anchor features capturing global dynamics and decodes velocity information at arbitrary timestamps.
Related event: Uni4R Unifies 4D Reconstruction and Tracking via OT and ODE(2 posts)→
More from Research
- GLINT (CVPR 2026 Oral): Solving Transparency in 3D Gaussian Splatting via Radiance Decomposition — rsasaki0109 · 2026-08-13
- Anthropic's New Paper: Reading Claude's Internal Thoughts in Plain English — thisguyknowsai · 2026-08-13
- Tencent's $850K KDD Cup: Industrial Track Champion Coded Entirely with DeepSeek — 赛博禅心 · 2026-08-13
- Quantifying and Training 'Taste' as LLMs Near Self-Improvement — Kangwook_Lee · 2026-08-13
- François Chollet: Test-Time Training Shows Great Potential to Go Mainstream — fchollet · 2026-08-13
- DeWorldSG (ECCV 2026): Generating 3D Semantic Scene Graphs via World-Model Priors — rsasaki0109 · 2026-08-13