Research map / Video / Video segmentation and tracking
Particle Video Revisited: Tracking through Occlusions Using Point Trajectories
Adam W. Harley, Zhaoyuan Fang, Katerina Fragkiadaki
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 PIPs multi-frame trajectories instead of chained RAFT flow | RAFT · FlyingThings++ test (occluded targets) | trajectory error: 46.73 -> 36.67 | Improved |
| use PIPs instead of RAFT for visible targets | RAFT · FlyingThings++ test (visible targets) | trajectory error: 24.32 -> 15.54 | Improved |
| use PIPs instead of RAFT on KITTI | RAFT · KITTI trajectories | trajectory error: visible 4.03 -> 4.40, occluded 6.79 -> 5.56 | Mixed |
| use PIPs instead of RAFT on CroHD | RAFT · CroHD trajectories | trajectory error: visible 7.91 -> 5.16, occluded 13.04 -> 7.56 | Improved |
| use PIPs instead of DINO on BADJA | DINO · BADJA keypoint propagation (average) | PCK-T: 50.5 -> 62.3 | Improved |
| add spatial windowing to DINO (ablation) | DINO · BADJA keypoint propagation (average) | PCK-T: 50.5 -> 53.5 | Improved |
| add spatial windowing to RAFT (ablation) | RAFT · BADJA keypoint propagation (average) | PCK-T: 45.6 -> 43.5 | Hurt |
| add spatial windowing to ImageNet ResNet (ablation) | ImageNet ResNet · BADJA keypoint propagation (average) | PCK-T: 34.4 -> 44.6 | Improved |
| retrain RAFT on FlyingThings++ occlusion data (ablation) | RAFT · FlyingThings++ test | trajectory error | No clear effect |
| train PIPs with T=32 instead of T=8 (ablation) | PIPs · FlyingThings++ | training difficulty | Hurt |
| use more than 6 update iterations (ablation) | PIPs · FlyingThings++ | trajectory error | No clear effect |
| use test-time stride 4 instead of training stride 8 (ablation) | PIPs · high-resolution test sets | trajectory error | Improved |
Closest papers by meaning
- Dense Optical Tracking: Connecting the Dots (CVPR 2024 · similarity 0.92)
- FlowTrack: Revisiting Optical Flow for Long-Range Dense Tracking (CVPR 2024 · similarity 0.90)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · similarity 0.89)
- Tracking Everything Everywhere across Multiple Cameras (AAAI 2025 · similarity 0.88)
- SpatialTracker: Tracking Any 2D Pixels in 3D Space (CVPR 2024 · similarity 0.88)
- MotionPro: A Precise Motion Controller for Image-to-Video Generation (CVPR 2025 · similarity 0.87)
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.
