Research map / 3D vision / Depth and stereo
PointOdyssey: A Large-Scale Synthetic Dataset for Long-Term Point Tracking
Yang Zheng, Adam W. Harley, Bokui Shen, Gordon Wetzstein, Leonidas J. Guibas
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 |
|---|---|---|---|
| widen temporal window by replacing MLP-Mixer with 1D ResNet (PIPs+) (ablation) | PIPs · PointOdyssey test | median trajectory error: 63.98 -> 28.93 | Improved |
| replace MLP-Mixer with 1D ResNet (ablation) | PIPs · inference speed on long videos | FPS: 4 FPS -> 55 FPS | Improved |
| add multi-template strategy (PIPs+ to PIPs++) (ablation) | PIPs+ · PointOdyssey test | median trajectory error: 28.93 -> 26.95 | Improved |
| add multi-template strategy (PIPs+ to PIPs++) (ablation) | PIPs+ · CroHD | delta_avg: 45.51% -> 44.09% | Hurt |
| retrain PIPs from scratch on PointOdyssey instead of original weights | PIPs · TAP-Vid-DAVIS | median trajectory error: 6.30 -> 5.14 | Improved |
| retrain PIPs from scratch on PointOdyssey instead of original weights | PIPs · CroHD | median trajectory error: 19.23 -> 11.94 | Improved |
| retrain TAP-Net on PointOdyssey with limited compute instead of authors' Kubric weights | TAP-Net · TAP-Vid-DAVIS | median trajectory error: 10.56 -> 36.19 | Hurt |
| drop visibility classification and score losses when training PIPs+/PIPs++ (ablation) | PIPs++ · tracking | tracking performance | No clear effect |
| use 36-frame non-overlapping windows at test time (ablation) | PIPs++ · inference window length | tracking performance | Improved |
Closest papers by meaning
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.89)
- DriveTrack: A Benchmark for Long-Range Point Tracking in Real-World Videos (CVPR 2024 · similarity 0.89)
- Direct Motion Models for Assessing Generated Videos (ICML 2025 · similarity 0.87)
- TAPVid-3D: A Benchmark for Tracking Any Point in 3D (NeurIPS 2024 · similarity 0.87)
- Fast Encoder-Based 3D from Casual Videos via Point Track Processing (NeurIPS 2024 · similarity 0.86)
- Tracktention: Leveraging Point Tracking to Attend Videos Faster and Better (CVPR 2025 · similarity 0.86)
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.
