Research map / Architectures and efficiency / CNN design
SwinTrack: A Simple and Strong Baseline for Transformer Tracking
Liting Lin, Heng Fan, Zhipeng Zhang, Yong Xu, Haibin Ling
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 |
|---|---|---|---|
| replace ResNet-50 backbone with Swin Transformer backbone (ablation) | SwinTrack-T-224 (no motion token) · LaSOT_ext | SUC: 41.8 -> 46.9 | Improved |
| replace ResNet-50 backbone with Swin Transformer backbone (ablation) | SwinTrack-T-224 (no motion token) · LaSOT | SUC: 64.2 -> 66.7 | Improved |
| concatenation-based fusion instead of cross-attention-based fusion (ablation) | SwinTrack-T-224 · LaSOT_ext | SUC: 45.4 -> 46.9 | Improved |
| concatenation-based fusion instead of cross-attention-based fusion (ablation) | SwinTrack-T-224 · LaSOT | SUC: 66.6 -> 66.7 | No clear effect |
| encoder-only with search-token decoder instead of target-query decoder (ablation) | SwinTrack-T-224 · LaSOT_ext | SUC: 43.2 -> 46.9 | Improved |
| untied positional encoding instead of absolute sine encoding (ablation) | SwinTrack-T-224 · LaSOT | SUC: 65.7 -> 66.7 | Improved |
| varifocal IoU-aware classification loss instead of BCE (ablation) | SwinTrack-T-224 · TrackingNet | SUC: 79.4 -> 80.8 | Improved |
| keep Hanning penalty window in post-processing (ablation) | SwinTrack-T-224 · LaSOT | SUC: 65.7 -> 66.7 | Improved |
| add motion token (ablation) | SwinTrack-T-224 · GOT-10k | mAO: 70.0 -> 71.3 | Improved |
| add motion token (ablation) | SwinTrack-B-384 · TrackingNet | SUC: 84.0 -> 84.0 | No clear effect |
| add motion token to SwinTrack-B-384 (ablation) | SwinTrack-B-384 · LaSOT | SUC: 70.2 -> 71.3 | Improved |
| replace motion token with a learnable embedding token (ablation) | SwinTrack-T-224 · LaSOT_ext | SUC: 47.6 -> 45.2 | Hurt |
Closest papers by meaning
- AiATrack: Attention in Attention for Transformer Visual Tracking (ECCV 2022 · similarity 0.90)
- High-Performance Transformer Tracking (TPAMI 2022 · similarity 0.89)
- Correlation-Embedded Transformer Tracking: A Single-Branch Framework (TPAMI 2024 · similarity 0.88)
- Target Focused Shallow Transformer Framework for Efficient Visual Tracking (AAAI 2024 · similarity 0.86)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.85)
- MIMTrack: In-Context Tracking via Masked Image Modeling (AAAI 2025 · 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.
