Research map / Detection and segmentation / Tracking
TAP-Vid: A Benchmark for Tracking Any Point in a Video
Carl Doersch, Ankush Gupta, Larisa Markeeva, Adria Recasens Continente, Lucas Smaira, Yusuf Aytar, Joao Carreira, Andrew Zisserman, Yi Yang
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 |
|---|---|---|---|
| use TAP-Net instead of the strongest reported baseline, Kubric-VFS-Like | point tracking · TAP-Vid-Kinetics | AJ: 40.5 -> 46.6 | Improved |
| use TAP-Net instead of PIPs | point tracking · TAP-Vid-Kubric | AJ: 59.1 -> 65.4 | Improved |
| use TAP-Net instead of PIPs | point tracking · TAP-Vid-DAVIS | AJ: 42.0 -> 38.4 | Hurt |
| use TAP-Net instead of Kubric-VFS-Like | point tracking · TAP-Vid-RGB-Stacking | AJ: 57.9 -> 59.9 | Improved |
| add limited time-shifting in the first two backbone layers (ablation) | TAP-Net · TAP-Vid-Kinetics | AJ: 46.0 -> 46.6 | Improved |
| use time-shifting only in the first two layers rather than every layer (ablation) | TAP-Net · TAP-Vid-Kinetics | AJ: 45.4 -> 46.6 | Improved |
| use soft argmax instead of a 2-layer MLP regressor (ablation) | TAP-Net · TAP-Vid-Kinetics | AJ: 45.8 -> 46.6 | Improved |
| sample queries at a stride of 5 frames rather than only at the first visible frame (ablation) | TAP-Net · TAP-Vid-Kinetics | AJ: 38.5 -> 46.6 | Improved |
| sample queries at a stride of 5 frames rather than only at the first visible frame (ablation) | TAP-Net · TAP-Vid-DAVIS | AJ: 33.0 -> 38.4 | Improved |
| sample queries at a stride of 5 frames rather than only at the first visible frame (ablation) | TAP-Net · TAP-Vid-RGB-Stacking | AJ: 53.5 -> 59.9 | Improved |
| add optical-flow track assistance to human annotation (ablation) | annotation pipeline · Kubric simulated annotation | AJ: 74.7 -> 81.4 | Improved |
| fine-tune the Kubric-trained model on Kinetics | TAP-Net · JHMDB val | PCK@0.1: 62.3 -> 63.4 | Improved |
Closest papers by meaning
- TAPVid-3D: A Benchmark for Tracking Any Point in 3D (NeurIPS 2024 · similarity 0.94)
- TAPIP3D: Tracking Any Point in Persistent 3D Geometry (NeurIPS 2025 · similarity 0.90)
- TAPVid-360: Tracking Any Point in 360 from Narrow Field of View Video (NeurIPS 2025 · similarity 0.89)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · similarity 0.89)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.89)
- ETAP: Event-based Tracking of Any Point (CVPR 2025 · similarity 0.89)
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.
