Research map / Detection and segmentation / Tracking
Tracking by Natural Language Specification with Long Short-term Context Decoupling
Ding Ma, Xiangqian 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 both SCM and LCP branches with LSM module to frame-token baseline (ablation) | DecoupleTNL · TNL2k (NL + Box) | AUC: 0.318 -> 0.567 | Improved |
| add short-term context-matching (SCM) branch (ablation) | frame-token baseline · LaSOT (NL + Box) | AUC: 0.427 -> 0.605 | Improved |
| add long-term context-perceiving (LCP) branch (ablation) | frame-token baseline · LaSOT (NL + Box) | AUC: 0.427 -> 0.620 | Improved |
| use LSM module to fuse long and short-term tokens instead of sum, concat or multi-conv (ablation) | DecoupleTNL · TNL2k (NL + Box) | AUC: 52.9 -> 56.7 (vs sum) | Improved |
| use short-term temporal scope T=5 rather than 2, 3, 4 or 6 (ablation) | SCM branch · TNL2k (NL + Box) | AUC: 56.7 vs 56.4, 55.2, 53.8, 56.1 | Mixed |
| use sigmoid or sigmoid+multi-conv for frame tokens instead of scale-and-shift modulation (ablation) | LSM module · TNL2k (NL + Box) | AUC: 56.7 -> 55.1 (sigmoid), 55.6 (sigmoid+multi-conv) | Hurt |
| stack 4 LSM modules versus 1 to 5 (ablation) | LSM module · TNL2k (NL + Box) | AUC: 56.7 at 4 stacks; 55.7 to 56.6 for others | No clear effect |
| use contrastive (InfoNCE) learning for the context branches instead of token concatenation (ablation) | DecoupleTNL · TNL2k (NL + Box) | AUC and training time | Improved |
| add long short-term context decoupling strategy to existing TNL trackers (TNL, SNLT, CapsuleTNL, AdaSwitcher) | existing TNL trackers · TNL2k | AUC | Improved |
| give baseline trackers the same ResNet50 + transformer backbone (ablation) | ToMP, TransT, AiATrack, TrDiMP, SNLT, AdaSwitcher · OTB-lang and LaSOT | AUC | No clear effect |
| initialize with NL only instead of NL + Box (ablation) | DecoupleTNL · TNL2k | AUC: 0.567 -> 0.407 | Hurt |
Closest papers by meaning
- Beyond Explicit Language: Plug-and-Play Visual-to-Linguistic Modeling Toward General Object Tracking (CVPR 2026 · similarity 0.90)
- ChatTracker: Enhancing Visual Tracking Performance via Chatting with Multimodal Large Language Model (NeurIPS 2024 · similarity 0.87)
- Unifying Visual and Vision-Language Tracking via Contrastive Learning (AAAI 2024 · similarity 0.87)
- Less Is More: Token Context-Aware Learning for Object Tracking (AAAI 2025 · similarity 0.87)
- Boundary-Aware Temporal Dynamic Pseudo-Supervision Pairs Generation for Zero-Shot Natural Language Video Localization (AAAI 2025 · similarity 0.86)
- ReVisionLLM: Recursive Vision-Language Model for Temporal Grounding in Hour-Long Videos (CVPR 2025 · 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.
