Research map / Label-efficient and robust learning / Knowledge distillation
General Compression Framework for Efficient Transformer Object Tracking
Lingyi Hong, Jinglun Li, Xinyu Zhou, Shilin Yan, Pinxue Guo, Kaixun Jiang, Zhaoyu Chen, Shuyong Gao, Runze Li, Xingdong Sheng, Wei Zhang, Hong Lu and 1 more
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 prediction guidance (ablation) | CompressTracker-4 · LaSOT (CompressTracker-4, OSTrack teacher) | AUC: 62.8 -> 63.5 | Improved |
| add feature mimicking (ablation) | CompressTracker-4 · LaSOT (CompressTracker-4, OSTrack teacher) | AUC: 62.8 -> 63.3 | Improved |
| add replacement training (ablation) | CompressTracker-4 · LaSOT (CompressTracker-4, OSTrack teacher) | AUC: 62.8 -> 63.7 | Improved |
| combine prediction guidance, feature mimicking and replacement training (ablation) | CompressTracker-4 · LaSOT (CompressTracker-4, OSTrack teacher) | AUC: 62.8 -> 65.2 | Improved |
| use random stage replacement instead of decoupled stage training (ablation) | CompressTracker-4 · LaSOT (CompressTracker-4, OSTrack teacher) | AUC: 64.6 -> 65.2 | Improved |
| add progressive replacement (ablation) | CompressTracker-4 · LaSOT (CompressTracker-4, OSTrack teacher) | AUC: 64.8 -> 65.2 | Improved |
| train 500 epochs instead of 300 (ablation) | CompressTracker-4 · LaSOT (CompressTracker-4, OSTrack teacher) | AUC: 65.2 -> 66.1 | Improved |
| initialize decoder from teacher instead of random (ablation) | CompressTracker-4 · LaSOT | AUC: 62.3 -> 62.8 | Improved |
| keep decoder trainable instead of frozen (ablation) | CompressTracker-4 · LaSOT | AUC: 62.6 -> 62.8 | Improved |
| initialize backbone by skipping layers of OSTrack instead of first 4 layers of MAE-B (ablation) | CompressTracker-4 · LaSOT | AUC: 59.9% -> 62.3% | Improved |
| use even stage division instead of uneven (2,2,6,2) (ablation) | CompressTracker-4 · LaSOT | AUC: 62.7% -> 62.8% | No clear effect |
| use CompressTracker instead of plain distillation | CompressTracker-4 · LaSOT | AUC: 63.8 -> 66.1 | Improved |
Closest papers by meaning
- Generalized Relation Modeling for Transformer Tracking (CVPR 2023 · similarity 0.87)
- Compact Transformer Tracker with Correlative Masked Modeling (AAAI 2023 · similarity 0.86)
- Exploring the Feature Extraction and Relation Modeling For Light-Weight Transformer Tracking (ECCV 2024 · similarity 0.86)
- MixFormerV2: Efficient Fully Transformer Tracking (NeurIPS 2023 · similarity 0.86)
- TransCenter: Transformers With Dense Representations for Multiple-Object Tracking (TPAMI 2022 · similarity 0.86)
- Target Focused Shallow Transformer Framework for Efficient Visual Tracking (AAAI 2024 · similarity 0.84)
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.
