Research map / Label-efficient and robust learning / Semi-supervised learning
Real-World Point Tracking with Verifier-Guided Pseudo-Labeling
Görkay Aydemir, Fatma Güney, Weidi Xie
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 verifier-guided per-frame selection instead of random teacher selection (5 teachers) (ablation) | pseudo-label generation / inference ensemble · TAP-Vid DAVIS | delta_avg: 77.7 -> 81.1 | Improved |
| use verifier-guided per-frame selection instead of random teacher selection (5 teachers) (ablation) | pseudo-label generation / inference ensemble · RoboTAP | delta_avg: 78.0 -> 83.1 | Improved |
| add teacher D to teachers {A,B,C} (ablation) | verifier · TAP-Vid DAVIS | delta_avg: 80.6 -> 80.8 | Improved |
| add more teachers (3 to 5) (ablation) | verifier · RoboTAP | delta_avg: 81.8 -> 83.1 | Improved |
| add synthetic data to real-only fine-tuning without schedule (ablation) | Track-On-R fine-tuning · EgoPoints | delta_avg: 66.9 -> 65.6 | Hurt |
| add synthetic data to real-only fine-tuning without schedule (ablation) | Track-On-R fine-tuning · EgoPoints | OA: 90.2 -> 90.7 | Improved |
| add loss-weight schedule increasing real-sample weight to mixed training (ablation) | Track-On-R fine-tuning · EgoPoints | delta_avg: 65.6 -> 67.3 | Improved |
| add schedule and synthetic mix compared with real-only (ablation) | Track-On-R fine-tuning · TAP-Vid DAVIS | delta_avg: 80.3 -> 80.3 | No clear effect |
| fine-tune Track-On2 with verifier-guided pseudo-labels on real videos | Track-On2 · EgoPoints | delta_avg: 61.7 -> 67.3 | Improved |
| fine-tune Track-On2 with verifier-guided pseudo-labels on real videos | Track-On2 · RoboTAP | AJ: 68.1 -> 70.9 | Improved |
| fine-tune Track-On2 with verifier-guided pseudo-labels on real videos | Track-On2 · TAP-Vid Kinetics | AJ: 55.3 -> 57.8 | Improved |
| fine-tune Track-On2 with verifier-guided pseudo-labels on real videos | Track-On2 · TAP-Vid DAVIS | AJ: 67.0 -> 68.1 | Improved |
Closest papers by meaning
- Track-On2: Enhancing Online Point Tracking With Memory (TPAMI 2026 · similarity 0.89)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.88)
- Tracktention: Leveraging Point Tracking to Attend Videos Faster and Better (CVPR 2025 · similarity 0.87)
- DriveTrack: A Benchmark for Long-Range Point Tracking in Real-World Videos (CVPR 2024 · similarity 0.85)
- Rethinking Pseudo-Label Guided Learning for Weakly Supervised Temporal Action Localization from the Perspective of Noise Correction (AAAI 2025 · similarity 0.85)
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 2025 · similarity 0.85)
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.
