Research map / Detection and segmentation / Tracking
TGTrack: Temporal Generative Learning for Unified Single Object Tracking
Wanting Geng, Xin Chen, Chuanyu Sun, Jie Zhao, Ben Kang, Dong Wang, Huchuan Lu
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 the temporal generative decoder (ablation) | TGTrack-B256 · 5 benchmarks average (ablation) | average AUC change: -1.50 when removed | Improved |
| add the time token embedding (ablation) | TGTrack-B256 · 5 benchmarks average (ablation) | average AUC change: -0.44 when removed | Improved |
| add both decoder and time token embedding (ablation) | TGTrack-B256 · 5 benchmarks average (ablation) | average AUC change: -2.36 when both removed | Improved |
| add gated fusion module in the decoder (ablation) | TGTrack-B256 · 5 benchmarks average (ablation) | average AUC change: -0.84 when removed | Improved |
| make time token learnable (ablation) | TGTrack-B256 · 5 benchmarks average (ablation) | average AUC change: -0.58 when non-learnable | Improved |
| predict next frame instead of current frame (ablation) | TGTrack-B256 · 5 benchmarks average (ablation) | average AUC change: -0.90 when predicting current frame | Improved |
| vary generative decoder depth (4, 6, 10 vs 8 blocks) (ablation) | TGTrack-B256 · 5 benchmarks average (ablation) | average AUC change: -0.32 (4), -0.08 (6), -0.20 (10) | Mixed |
| vary video clip length (3, 4, 6 vs 5 frames) (ablation) | TGTrack-B256 · 5 benchmarks average (ablation) | average AUC change: -0.36 (3), -0.10 (4), -0.06 (6) | Mixed |
| vary generative loss weight (0.01, 0.5, 1.0 vs 0.1) (ablation) | TGTrack-B256 · 5 benchmarks average (ablation) | average AUC change: -0.08 (0.01), -0.06 (0.5), -0.20 (1.0) | Mixed |
| temporal generative learning in unified tracker (vs SUTrack-B384) | TGTrack-B384 · LaSOT | AUC: 74.4 -> 75.3 | Improved |
| scale backbone from B384 to L384 | TGTrack · LaSOT | AUC: 75.3 -> 76.4 | Improved |
| use generative decoder only in training (disabled at inference) | TGTrack · inference cost | overhead | No clear effect |
Closest papers by meaning
- ODTrack: Online Dense Temporal Token Learning for Visual Tracking (AAAI 2024 · similarity 0.89)
- MIMTrack: In-Context Tracking via Masked Image Modeling (AAAI 2025 · similarity 0.88)
- Explicit Visual Prompts for Visual Object Tracking (AAAI 2024 · similarity 0.87)
- UTPTrack: Towards Simple and Unified Token Pruning for Visual Tracking (CVPR 2026 · similarity 0.87)
- SPMTrack: Spatio-Temporal Parameter-Efficient Fine-Tuning with Mixture of Experts for Scalable Visual Tracking (CVPR 2025 · similarity 0.87)
- Autoregressive Sequential Pretraining for Visual Tracking (CVPR 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.
