Research map / Architectures and efficiency / Attention and transformers
Target-Aware Tracking with Long-Term Context Attention
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 LCA layers to the backbone instead of using only Swin Transformer stages (ablation) | TATrack-S · GOT-10k | AO: 72.9 -> 74.3 | Helped |
| Add image-relationship location encodings to LCA (ablation) | TATrack-S · GOT-10k | AO: 71.3 -> 74.3 | Helped |
| Add ltrb box embedding (ablation) | TATrack-S · GOT-10k | AO: 73.9 -> 74.3 | Helped |
| Add ltrb box embedding (ablation) | TATrack-S · TrackingNet | AUC: 82.0 -> 81.8 | Hurt |
| Add Gaussian localization embedding (ablation) | TATrack-S · LaSOT | AUC: 66.9 -> 68.1 | Helped |
| Include background in the previous template rather than using only the target (ablation) | TATrack-S · TrackingNet | AUC: 80.8 -> 81.8 | Helped |
| Include background in both the target and previous templates (ablation) | TATrack-S · TrackingNet | AUC: 81.8 -> 82.2 | Helped |
| Update the previous template with the last frame instead of using the initial frame (ablation) | TATrack-S · LaSOT | AUC: 66.7 -> 61.0 | Hurt |
| Use historical mean confidence as the dynamic update threshold instead of no updates (ablation) | TATrack-S · GOT-10k | AO: 71.4 -> 74.3 | Helped |
| Use historical mean confidence as the dynamic update threshold instead of no updates (ablation) | TATrack-S · LaSOT | AUC: 66.7 -> 66.1 | Hurt |
| Use penalized historical mean confidence instead of ordinary historical mean confidence (ablation) | TATrack-S · LaSOT | AUC: 66.1 -> 68.1 | Helped |
| Use penalized historical mean confidence instead of ordinary historical mean confidence (ablation) | TATrack-S · GOT-10k | AO: 74.3 -> 74.0 | Hurt |
Closest papers by meaning
- GlobalTrack: A Simple and Strong Baseline for Long-Term Tracking (AAAI 2020 · similarity 0.89)
- Less Is More: Token Context-Aware Learning for Object Tracking (AAAI 2025 · similarity 0.88)
- MambaLCT: Boosting Tracking via Long-term Context State Space Model (AAAI 2025 · similarity 0.88)
- Attentional Correlation Filter Network for Adaptive Visual Tracking (CVPR 2017 · similarity 0.88)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.87)
- HIPTrack: Visual Tracking with Historical Prompts (CVPR 2024 · 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.