Research map / Detection and segmentation / Open-set and few-shot detection
SAM2-OV: A Novel Detection-Only Tuning Paradigm for Open-Vocabulary Multi-Object Tracking
Yangkai Chen, Qiangqiang Wu, Guangyao Li, Junlong Gao, Guanglin Niu, Hanzi Wang
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 SEA distillation adapter to UDM (ablation) | SAM2-OV · TAO val | TETA: 38.5 -> 39.5 | Improved |
| add MPAM to Base (without SEA) (ablation) | SAM2-OV · TAO val | TETA: 38.5 -> 40.8 | Improved |
| add MPAM on top of Base+SEA (ablation) | SAM2-OV · TAO val | TETA: 39.5 -> 39.6 | Mixed |
| add MPAM to Base (without SEA) (ablation) | SAM2-OV · TAO val | novel ClsA: 2.4 -> 2.1 | Hurt |
| use UDM detection prompts instead of grid prompts for SAM2 (ablation) | SAM2 · TAO val | TETA: 16.5 -> 38.5 | Improved |
| raise SEA distillation weight lambda4 from 2 to 3 (ablation) | SAM2-OV · TAO val | TETA: 40.1 -> 39.2 | Hurt |
| raise SEA distillation weight lambda4 from 1 to 2 (ablation) | SAM2-OV · TAO val | novel ClsA: 5.1 -> 7.5 | Improved |
| increase detection cap per frame from 30 to 45 (ablation) | SAM2-OV · TAO val | TETA: 39.4 -> 39.6 | Improved |
| increase detection cap per frame from 45 to 60 (ablation) | SAM2-OV · TAO val | TETA: 39.6 -> 39.2 | Hurt |
| prompt detection every frame instead of every 5 (ablation) | SAM2-OV · TAO val | TETA: 39.6 -> 39.7 | No clear effect |
| lengthen detection interval from 5 to 7 frames (ablation) | SAM2-OV · TAO val | TETA: 39.6 -> 39.2 | Hurt |
| raise MPAM motion weight from 0.1 to 0.5 (ablation) | SAM2-OV · TAO val | TETA: 39.6 -> 38.9 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- OVTR — OVTR: End-to-End Open-Vocabulary Multiple Object Tracking with Transformer (ICLR 2025) · better on 4 of 5 shared results
- OVTrack — OVTrack: Open-Vocabulary Multiple Object Tracking (CVPR 2023) · better on 0 of 1 shared result
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
- COVTrack++: Learning Open-Vocabulary Multi-Object Tracking from Continuous Videos via a Synergistic Paradigm (TPAMI 2026 · similarity 0.93)
- OVT-B: A New Large-Scale Benchmark for Open-Vocabulary Multi-Object Tracking (NeurIPS 2024 · similarity 0.90)
- OVTR: End-to-End Open-Vocabulary Multiple Object Tracking with Transformer (ICLR 2025 · similarity 0.89)
- Multiple Object Tracking as ID Prediction (CVPR 2025 · similarity 0.89)
- Focusing on Tracks for Online Multi-Object Tracking (CVPR 2025 · similarity 0.88)
- S2-Track: A Simple yet Strong Approach for End-to-End 3D Multi-Object Tracking (ICML 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.
