Research map / Video / Video segmentation and tracking
DINO-Tracker: Taming DINO for Self-Supervised Point Tracking in a Single Video
Narek Tumanyan, Assaf Singer, Shai Bagon, Tali Dekel
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 pretrained DINOv2 features (vs training CNN from scratch without DINO) (ablation) | DINO-Tracker · TAP-Vid-DAVIS-480 | delta_avg^x: 71.4 -> 80.4 | Improved |
| predict residual with CNN Delta-DINO instead of LoRA fine-tuning DINOv2 (ablation) | DINO-Tracker · TAP-Vid-DAVIS-480 | delta_avg^x: 73.2 -> 80.4 | Improved |
| add DINO best-buddies loss (ablation) | DINO-Tracker · TAP-Vid-DAVIS-480 | delta_avg^x: 78.2 -> 80.4 | Improved |
| add prior preservation loss (ablation) | DINO-Tracker · TAP-Vid-DAVIS-480 | delta_avg^x: 79.2 -> 80.4 | Improved |
| add optical flow loss (ablation) | DINO-Tracker · TAP-Vid-DAVIS-480 | delta_avg^x: 78.3 -> 80.4 | Improved |
| add refined best-buddies and cycle-consistency losses (ablation) | DINO-Tracker · TAP-Vid-DAVIS-480 | AJ: 63.2 -> 64.6 | Improved |
| refine DINO features with test-time training vs raw DINOv2 matching | DINOv2 · TAP-Vid-DAVIS-480 | delta_avg^x: 64.7 -> 80.4 | Improved |
| use DINOv2 layer 16 tokens rather than other layers/facets (ablation) | raw DINOv2 tracking · TAP-Vid-DAVIS-480 | delta_avg^x: 59.9 -> 64.7 | Improved |
| train only on 50% of frames (every 2nd frame) (ablation) | DINO-Tracker · TAP-Vid-DAVIS-480 | delta_avg^x: 80.7 -> 80.6 | No clear effect |
| train only on 25% of frames (every 4th frame) (ablation) | DINO-Tracker · TAP-Vid-DAVIS-480 | delta_avg^x: 80.7 -> 79.1 | Hurt |
| combine test-time training with DINO prior vs Omnimotion | DINO-Tracker · TAP-Vid-DAVIS-480 | delta_avg^x: 74.1 -> 80.4 | Improved |
Closest papers by meaning
- Track-On2: Enhancing Online Point Tracking With Memory (TPAMI 2026 · similarity 0.89)
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 2025 · similarity 0.88)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.87)
- Exploring Temporally-Aware Features for Point Tracking (CVPR 2025 · similarity 0.87)
- ODTrack: Online Dense Temporal Token Learning for Visual Tracking (AAAI 2024 · similarity 0.86)
- DriveTrack: A Benchmark for Long-Range Point Tracking in Real-World Videos (CVPR 2024 · 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.
