ReLift Hits SOTA Dense 3D Tracking on WorldTrack With 2.7x Faster Pipeline
Technion · hf · 2026-11-02
- Paper: ReLift: Dense 3D Tracking as Lifted Trajectory Repair, from a Technion–NVIDIA team (co-first authors Assaf Singer and Maroun Khriesh; Or Litany among senior authors).
- Approach: Requires no image-based training. It combines pretrained 2D tracking with 3D geometry, repairing and completing lifted trajectories via learned geometric trajectory refinement to fix broken long-range 3D tracks.
- Results: State-of-the-art dense 3D tracking accuracy on WorldTrack (+8.8 APD over prior work), with a strong speed–accuracy trade-off — the full pipeline runs up to 2.7x the FPS of compared methods, clearly beating V-DPM, 4RC, TrackCraft3R, and DELTAv2 on the accuracy-vs-speed plot.
More from Research
- Microsoft and Berkeley open-source TokenEconomizer, cutting coding agent costs 18–31% — victor_explore · 2026-11-02
- TLA-Verus uses LLM agents to prove temporal properties of TLA+ specs — tianyin_xu · 2026-11-02
- J-Lens is just a first-order Taylor expansion, researcher points out — burny_tech · 2026-11-02
- Mollick: expert-level work benchmarks are brutally hard, and ARC-AGI, GDPval, METR are saturated — emollick · 2026-11-02
- Rebuilding a Protein Domain Distribution Pipeline After a Backup Failure — AryHHAry · 2026-11-02
- NVIDIA's Elastic Latent Recurrence lets one 4B model trade loop count for accuracy: 18→29 on AA Index — burny_tech · 2026-11-02