Research map / Vision-language and multimodal / Cross-modal learning
UniSOT: A Unified Framework for Multi-Modality Single Object Tracking
Yinchao Ma, Yuyang Tang, Wenfei Yang, Tianzhu Zhang, Zhou Yuan Xu, Feng He Wu
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 multi-modal contrastive loss (MMCLoss) (ablation) | UniSOT-B baseline · TNL2K (BBOX reference) | AUC: 57.3 -> 58.7 | Improved |
| add reference-adaptive box head (RABH) over anchor-free head (ablation) | UniSOT-B + MMCLoss · TNL2K (BBOX reference) | AUC: 58.7 -> 60.9 | Improved |
| add multi-modal contrastive loss (MMCLoss) (ablation) | UniSOT-B baseline · TNL2K (NL reference) | AUC: 49.6 -> 51.9 | Improved |
| add reference-adaptive box head (RABH) over anchor-free head (ablation) | UniSOT-B + MMCLoss · TNL2K (NL reference) | AUC: 51.9 -> 53.9 | Improved |
| add multi-modal contrastive loss (MMCLoss) (ablation) | UniSOT-B baseline · TNL2K (NL+BBOX reference) | AUC: 57.4 -> 59.8 | Improved |
| add reference-adaptive box head (RABH) over anchor-free head (ablation) | UniSOT-B + MMCLoss · TNL2K (NL+BBOX reference) | AUC: 59.8 -> 61.2 | Improved |
| use 1 fusion layer (N=11, M=1) instead of 6 fusion layers (N=6, M=6) (ablation) | UniSOT-B · TNL2K (NL reference) | AUC: 53.9 -> 45.1 | Hurt |
| use 6 separate + 6 fusion layers instead of 0 separate layers (ablation) | UniSOT-B · TNL2K (BBOX reference) | AUC: 59.7 -> 60.9 | Improved |
| use top-scoring negatives instead of random negatives (center positive) (ablation) | UniSOT-B · TNL2K (BBOX reference) | AUC: 59.9 -> 60.9 | Improved |
| update scenario tokens every 20 frames instead of never (ablation) | UniSOT-B · TNL2K (NL reference) | AUC: 52.4 -> 53.9 | Improved |
| initialize with BERT+MAE instead of MAE only (ablation) | UniSOT-B · TNL2K (NL reference) | AUC: 52.3 -> 53.9 | Improved |
| second stage training with auxiliary modality tuning blocks (AMTB) (ablation) | UniSOT-B* · DepthTrack | F-score: 53.8 -> 62.5 | Improved |
Closest papers by meaning
- Single-Model and Any-Modality for Video Object Tracking (CVPR 2024 · similarity 0.94)
- Collaborating Vision, Depth, and Thermal Signals for Multi-Modal Tracking: Dataset and Algorithm (NeurIPS 2025 · similarity 0.91)
- OneTracker: Unifying Visual Object Tracking with Foundation Models and Efficient Tuning (CVPR 2024 · similarity 0.88)
- UTPTrack: Towards Simple and Unified Token Pruning for Visual Tracking (CVPR 2026 · similarity 0.88)
- Bi-directional Adapter for Multimodal Tracking (AAAI 2024 · similarity 0.88)
- SUTrack: Towards Simple and Unified Single Object Tracking (AAAI 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.
