Research map / Detection and segmentation / Tracking
Exploring the Feature Extraction and Relation Modeling For Light-Weight Transformer Tracking
Jikai Zheng, Mingjiang Liang, Shaoli Huang, Jifeng Ning
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 |
|---|---|---|---|
| use cross-attention-only blocks in deep layers instead of mixed attention throughout (ablation) | FERMT · GOT-10k test | AO: 68.4 -> 68.6 | Improved |
| use A2A5 instead of full attention in all layers (ablation) | FERMT · GOT-10k test | attention computation: reduced by 42% and 31% versus A2 and A3 | Improved |
| remove feature extraction in all layers (cross-attention only) (ablation) | FERMT · GOT-10k test | AO: 68.6 -> 59.8 (A4), 63.2 (A5) | Hurt |
| use two-way instead of one-way interaction in cross blocks (ablation) | FERMT · GOT-10k test | AO: 67.3 -> 68.6 (A4 -> A5 with A2) | Improved |
| use asymmetric mixed attention (A2) in shallow layers (ablation) | FERMT · GOT-10k test | AO: 67.7 -> 68.6 (A1 -> A2 with A5) | Improved |
| layer ratio 6:6 instead of 3:9 (ablation) | FERMT · GOT-10k test | AO: 66.4 -> 68.6 | Improved |
| layer ratio 6:6 instead of 9:3 (ablation) | FERMT · GOT-10k test | AO: 68.3 -> 68.6 | Improved |
| add Dual Attention Unit preprocessing (ablation) | FERMT_A2A5 · GOT-10k test | AO: 68.6 -> 69.6 | Improved |
| add Dual Attention Unit preprocessing (ablation) | FERMT_A2A5 · UAV123 | AUC: 67.4 -> 67.5 | No clear effect |
Closest papers by meaning
- Correlation-Embedded Transformer Tracking: A Single-Branch Framework (TPAMI 2024 · similarity 0.89)
- Compact Transformer Tracker with Correlative Masked Modeling (AAAI 2023 · similarity 0.89)
- Generalized Relation Modeling for Transformer Tracking (CVPR 2023 · similarity 0.89)
- Unified Transformer Tracker for Object Tracking (CVPR 2022 · similarity 0.89)
- Target Focused Shallow Transformer Framework for Efficient Visual Tracking (AAAI 2024 · similarity 0.86)
- Is Multiple Object Tracking a Matter of Specialization? (NeurIPS 2024 · 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.
