Research map / Video / Video segmentation and tracking
Matching Every Pair to Track Every Point: PairFormer for All-Pairs Tracking and Video Trajectory Fields
Guangyang Wu, Youran Ding, Xinyu Che, Benyuan Sun, Yi Yang, Xiaohong 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 |
|---|---|---|---|
| allocate more depth to Broadcast Motion Mixer instead of the encoder (9/24 vs 24/9) (ablation) | PairFormer · APT-Bench held-out validation split | delta_avg: 66.9 -> 67.6 | Improved |
| allocate more depth to Broadcast Motion Mixer instead of the encoder (9/24 vs 24/9) (ablation) | PairFormer · APT-Bench held-out validation split | EPE: 4.4 -> 4.2 | Improved |
| add correspondence regularization Lcorr (ablation) | PairFormer · APT-Bench held-out validation split | delta_avg: 67.2 -> 67.8 | Improved |
| add correspondence regularization Lcorr (ablation) | PairFormer · APT-Bench held-out validation split | EPE: 4.3 -> 4.1 | Improved |
| increase CorrBank tokens from 32 to 64 (ablation) | PairFormer · APT-Bench held-out validation split | delta_avg: 67.1 -> 67.6 | Improved |
| increase CorrBank tokens from 64 to 128 (ablation) | PairFormer · APT-Bench held-out validation split | delta_avg: 67.6 -> 67.5 | No clear effect |
| use PairFormer all-pairs model instead of CoTracker3 Offline | TAP benchmarks · TAP benchmarks (8 datasets average) | delta_avg: 67.2 -> 68.1 | Improved |
| use PairFormer instead of CoTracker3 on DAVIS and DriveTrack | TAP benchmarks · TAPVid-DAVIS and DriveTrack | delta_avg: 77.1 -> 75.8 (DAVIS); 69.8 -> 66.9 (DriveTrack) | Hurt |
| use PairFormer instead of DOT on APT-Bench long videos | APT-Bench · APT-Bench | EPE: 3.68 -> 3.26 | Improved |
Closest papers by meaning
- Local All-Pair Correspondence for Point Tracking (ECCV 2024 · similarity 0.89)
- TTAPFormer: Robust Arbitrary Point Tracking via Transient Asynchronous Fusion of Frames and Events (CVPR 2026 · similarity 0.87)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.86)
- ETAP: Event-based Tracking of Any Point (CVPR 2025 · similarity 0.85)
- TAPVid-360: Tracking Any Point in 360 from Narrow Field of View Video (NeurIPS 2025 · similarity 0.84)
- TAPVid-3D: A Benchmark for Tracking Any Point in 3D (NeurIPS 2024 · similarity 0.84)
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.
