Research map / Detection and segmentation / Tracking
Track-On2: Enhancing Online Point Tracking With Memory
Görkay Aydemir, Weidi Xie, Fatma Güney
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 |
|---|---|---|---|
| add memory (ablation) | Track-On2 · TAP-Vid DAVIS | AJ: 56.2 -> 66.2 | Improved |
| add memory (ablation) | Track-On2 · RoboTAP | AJ: 61.9 -> 66.6 | Improved |
| add re-ranking module (ablation) | Track-On2 · TAP-Vid DAVIS | AJ: 61.5 -> 66.2 | Improved |
| add offset head (ablation) | Track-On2 · TAP-Vid DAVIS | delta_avg: 73.1 -> 79.1 | Improved |
| add offset head (ablation) | Track-On2 · TAP-Vid DAVIS | OA: 90.8 -> 91.5 | Hurt |
| replace Track-On training data/sampling with Track-On2 setup (ablation) | Track-On · RoboTAP | delta_avg: 76.4 -> 79.4 | Improved |
| increase training clip length T from 24 to 48 (ablation) | Track-On2 · PointOdyssey | Survival: 46.8 -> 51.5 | Improved |
| increase training memory size L from 18 to 24 (ablation) | Track-On2 · RoboTAP | delta_avg: 79.1 -> 78.9 | No clear effect |
| use DINOv3 ViT-S instead of DINOv2 ViT-S (ablation) | Track-On2 · TAP-Vid DAVIS | AJ: 66.2 -> 65.6 | Hurt |
| use DINOv3 ViT-S+ instead of ViT-S (ablation) | Track-On2 · RoboTAP | AJ: 66.4 -> 66.9 | Improved |
| unify dual memory into single memory with longer training (T=48, L=12) (ablation) | Track-On2 · RoboTAP | delta_avg: 80.0 -> 80.6 | No clear effect |
| sample uniformly with stride 2 instead of 1 (ablation) | Track-On2 · TAP-Vid DAVIS | AJ: 66.2 -> 66.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Track-On — Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025) · better on 13 of 13 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.97)
- Online Dense Point Tracking with Streaming Memory (ICCV 2025 · similarity 0.91)
- Tracktention: Leveraging Point Tracking to Attend Videos Faster and Better (CVPR 2025 · similarity 0.91)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · similarity 0.90)
- A Distractor-Aware Memory for Visual Object Tracking with SAM2 (CVPR 2025 · similarity 0.89)
- Is This Tracker On? A Benchmark Protocol for Dynamic Tracking (NeurIPS 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.
