Research map / 3D vision / Depth and stereo
Track4World: Feedforward World-centric Dense 3D Tracking of All Pixels
Jiahao Lu, Jiayi Xu, WENBO HU, Ruijie Zhu, Chengfeng Zhao, Sai Kit Yeung, Ying Shan, Yuan Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train with 2D flow supervision in addition to 3D data (ablation) | Track4World · scene flow decoder ablation (avg) | EPE3D: 0.6511 -> 0.2056 | Improved |
| add target lifting (lifted target samples) (ablation) | Track4World · scene flow decoder ablation (avg) | EPE3D: 0.3017 -> 0.2056 | Improved |
| use iterative refinement instead of single-step regression (ablation) | Track4World · scene flow decoder ablation (avg) | EPE3D: 0.2356 -> 0.2056 | Improved |
| add auxiliary 3D correlation and trajectory priors (ablation) | Track4World · scene flow decoder ablation (avg) | EPE3D: 0.2201 -> 0.2056 | Improved |
| use hybrid 2D flow plus Z-displacement instead of heuristic 2D flow + depth (ablation) | Track4World · scene flow decoder ablation (avg) | EPE3D: 0.8210 -> 0.2056 | Improved |
| use hybrid formulation instead of pure 3D flow regression (ablation) | Track4World · scene flow decoder ablation (avg) | EPE3D: 0.2815 -> 0.2056 | Improved |
| replace 2D-to-3D correlation with traditional 3D correlation (ablation) | Track4World · ADT dense tracking, 16 frames | memory / time: 3.4s -> OOM | Hurt |
| swap backbone initialization from MoGe to Pi3 to DA3 (ablation) | Track4World · avg scene flow | EPE3D: 0.3180 -> 0.2569 -> 0.2056 | Improved |
| Track4World vs best prior on world-coordinate 3D tracking | Track4World · 3D tracking world coords L-50 | APD: 0.4668 -> 0.5323 | Improved |
| Track4World vs CoTracker3 on 2D tracking | Track4World · TAP-Vid Kinetics | AJ: 55.8 -> 59.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- V-DPM — V-DPM: 4D Video Reconstruction with Dynamic Point Maps (CVPR 2026) · better on 2 of 2 shared results
- POMATO — POMATO: Marrying Pointmap Matching with Temporal Motions for Dynamic 3D Reconstruction (ICCV 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TrackingWorld: World-centric Monocular 3D Tracking of Almost All Pixels (NeurIPS 2025 · similarity 0.94)
- TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking (NeurIPS 2026 · similarity 0.91)
- TAPVid-3D: A Benchmark for Tracking Any Point in 3D (NeurIPS 2024 · similarity 0.89)
- DELTA: DENSE EFFICIENT LONG-RANGE 3D TRACKING FOR ANY VIDEO (ICLR 2025 · similarity 0.89)
- TAPIP3D: Tracking Any Point in Persistent 3D Geometry (NeurIPS 2025 · similarity 0.89)
- SpatialTracker: Tracking Any 2D Pixels in 3D Space (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
