Research map / Label-efficient and robust learning / Self-supervised and contrastive
Learning to LEAP: Efficient Dense Point Tracking by Focusing Where It Matters
Chenzhi Zhao, Wufan Wang, Bo Zhang, Wendong Wang
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 |
|---|---|---|---|
| add CSA to GMRW baseline (ablation) | GMRW · TAP-Vid Kubric (direct) | AJ: 51.4 -> 54.9 | Improved |
| add PkT to GMRW baseline (ablation) | GMRW · TAP-Vid Kubric (direct) | AJ: 51.4 -> 54.0 | Improved |
| add adaptive training scheme on top of CSA+PkT (ablation) | LEAP-Track · TAP-Vid Kubric (direct) | AJ: 55.3 -> 57.2 | Improved |
| use direct instead of chained evaluation (ablation) | LEAP-Track · TAP-Vid Kubric | AJ: 44.1 -> 57.2 | Improved |
| reduce evaluation stride from 2 to 1 (ablation) | LEAP-Track · TAP-Vid Kubric (direct) | AJ: 57.2 -> 58.6 | Improved |
| increase evaluation stride from 2 to 4 (ablation) | LEAP-Track · TAP-Vid Kubric (direct) | AJ: 57.2 -> 48.4 | Hurt |
| increase CSA key proportion beyond 75% (ablation) | LEAP-Track · TAP-Vid Kubric | AJ: slight degradation after peak at 75% | Hurt |
| increase PkT max neighbor size k beyond 64 (ablation) | LEAP-Track · TAP-Vid Kubric | AJ: degrades after peak at k = 64 | Hurt |
| sparse attention and k-NN walk in training (training throughput) | GMRW · Kinetics training | clips/s: 0.3923 -> 0.5806 | Improved |
| LEAP-Track vs GMRW direct on DAVIS | GMRW · TAP-Vid DAVIS (direct) | AJ: 30.3 -> 32.1 | Improved |
| LEAP-Track vs Flow-Walk on RGB-Stacking | Flow-Walk · TAP-Vid RGB-Stacking (direct) | AJ: 66.3 -> 58.8 | Hurt |
Closest papers by meaning
- TAPTRv2: Attention-based Position Update Improves Tracking Any Point (NeurIPS 2024 · similarity 0.88)
- 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)
- TAPIP3D: Tracking Any Point in Persistent 3D Geometry (NeurIPS 2025 · similarity 0.85)
- TAPVid-360: Tracking Any Point in 360 from Narrow Field of View Video (NeurIPS 2025 · similarity 0.85)
- 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.
