Research map / Detection and segmentation / Tracking
Correlation-Aware Deep Tracking
Check 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 |
|---|---|---|---|
| Replace baseline tracker features with SBT correlation-aware features | SiamFCpp on GOT-10k · GOT-10k test | AO: 59.5 -> 64.7 | Helped |
| Replace baseline tracker features with SBT correlation-aware features | DiMP on GOT-10k · GOT-10k test | AO: 61.1 -> 65.2 | Helped |
| Replace baseline tracker features with SBT features | STARK on GOT-10k · GOT-10k test | AO: 67.2 -> 68.3 | Helped |
| Replace ResNet-50 features with SBT features | STM on DAVIS17 · DAVIS17 | Mean J/F: 71.6 -> 74.2 | Helped |
| Embed correlation in the feature extraction pipeline rather than doing layer-wise correlation (ablation) | SBT-base on GOT-10k · GOT-10k test | AO: 60.1 -> 65.9 | Helped |
| Use CA instead of depth-wise correlation (ablation) | SBT-base feature correlation · GOT-10k test | AO: 60.1 -> 61.5 | Helped |
| Use a hierarchical multi-stage architecture instead of a single stage (ablation) | SBT · GOT-10k test | AO: 47.5 -> 63.7 | Helped |
| Replace conditional positional encoding with relative positional encoding (ablation) | SBT A5 · GOT-10k test | AO: 63.5 -> 63.1 | No clear effect |
| Use pure SRG blocks rather than mixing local self-attention and SRG cross-attention (ablation) | SBT · GOT-10k test | AO: 60.1 -> 63.5 | Helped |
| Move the earliest cross-attention block earlier (ablation) | SBT-base · GOT-10k test | AO: 63.2 -> 65.7 | Helped |
| Use an interleaved EoC-SA/EoC-CA pattern instead of a Siamese-style or separated pattern (ablation) | SBT-base · GOT-10k test | AO: 61.0 -> 65.0 | Helped |
| Increase third-stage channel size to 512 while keeping shallow dimensions at [64, 128] (ablation) | SBT-base · GOT-10k test | AO: 63.6 -> 61.8 | Hurt |
Closest papers by meaning
- Target-Aware Deep Tracking (CVPR 2019 · similarity 0.91)
- Graph Attention Tracking (CVPR 2021 · similarity 0.90)
- Attentional Correlation Filter Network for Adaptive Visual Tracking (CVPR 2017 · similarity 0.90)
- Learning Attentions: Residual Attentional Siamese Network for High Performance Online Visual Tracking (CVPR 2018 · similarity 0.90)
- Deformable Siamese Attention Networks for Visual Object Tracking (CVPR 2020 · similarity 0.89)
- Deeper and Wider Siamese Networks for Real-Time Visual Tracking (CVPR 2019 · similarity 0.89)
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.