Research map / Architectures and efficiency / Attention and transformers
AiATrack: Attention in Attention for Transformer Visual Tracking
Shenyuan Gao, Chunluan Zhou, Chao Ma, Xinggang Wang, Junsong Yuan
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 AiA in both self- and cross-attention (ablation) | AiATrack framework · LaSOT | AUC: 67.0 -> 68.7 | Improved |
| add AiA in both self- and cross-attention (ablation) | AiATrack framework · LaSOT-Ext | AUC: 44.7 -> 46.8 | Improved |
| add AiA only in self-attention (ablation) | AiATrack framework · LaSOT | AUC: 67.0 -> 68.6 | Improved |
| add AiA only in cross-attention (ablation) | AiATrack framework · LaSOT | AUC: 67.0 -> 67.5 | Improved |
| add positional encoding in AiA (ablation) | AiATrack · LaSOT | AUC: 68.0 -> 68.7 | Improved |
| add target-background embeddings instead of masking background (ablation) | AiATrack framework · LaSOT | AUC: 64.3 -> 67.0 | Improved |
| add target-background embeddings vs plain features (ablation) | AiATrack framework · LaSOT | AUC: 65.8 -> 67.0 | Improved |
| mask out background reference features (ablation) | AiATrack framework · LaSOT | AUC: 65.8 -> 64.3 | Hurt |
| add short-term reference branch (ablation) | AiATrack framework · LaSOT | AUC: 66.5 -> 67.0 | Improved |
| split into long-term and short-term branches (ablation) | AiATrack framework · LaSOT | AUC: 63.8 -> 67.0 | Improved |
| double attention blocks (cascade) instead of adding AiA (ablation) | AiATrack framework · LaSOT | AUC: 67.0 -> 67.1 | No clear effect |
| replace inner attention with convolutional bottleneck (ablation) | AiATrack · LaSOT | AUC: 68.7 -> 67.9 | Hurt |
Closest papers by meaning
- High-Performance Transformer Tracking (TPAMI 2022 · similarity 0.91)
- Transformer Tracking (CVPR 2021 · similarity 0.91)
- Transformer Tracking With Cyclic Shifting Window Attention (CVPR 2022 · similarity 0.90)
- Target Focused Shallow Transformer Framework for Efficient Visual Tracking (AAAI 2024 · similarity 0.88)
- Cross-modulated Attention Transformer for RGBT Tracking (AAAI 2025 · similarity 0.86)
- UTPTrack: Towards Simple and Unified Token Pruning for Visual Tracking (CVPR 2026 · similarity 0.85)
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.
