Research map / Detection and segmentation / Tracking
Towards Universal Physical Attacks on Single Object Tracking
Li Ding, Yongwei Wang, Kaiwen Yuan, Minyang Jiang, Ping Wang, Hua Huang, Z. Jane Wang
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 a universal adversarial patch with MTD and dilation shape loss | Siamese visual trackers · LaSOT person, car, and bottle | tracking performance drop: Person Success: 8.4, 42.5; Car Success: 46.9, 57.5; Bottle Success: 49.5, 38.2 | Improved |
| Use a universal adversarial patch with MTD and shrinking shape loss | Siamese visual trackers · LaSOT person, car, and bottle | tracking performance drop: Person Success: 38.8, 65.1; Car Success: 32.2, 44.5; Bottle Success: 45.7, 25.4 | Mixed |
| Add MTD loss to the dilation attack (ablation) | SiamMask · LaSOT person | success performance drop: 28.5 -> 37.0 | Improved |
| Add MTD loss to the shrinking attack (ablation) | SiamMask · LaSOT person | success performance drop: 53.7 -> 65.1 | Improved |
| Add MTD loss to the dilation attack (ablation) | SiamMask · LaSOT person | precision performance drop: 26.5 -> 35.8 | Improved |
| Add MTD loss to the shrinking attack (ablation) | SiamMask · LaSOT person | precision performance drop: 55.5 -> 74.2 | Improved |
| Add MTD loss to the dilation attack (ablation) | SiamRPN++ · LaSOT person | success performance drop: 31.2 -> 42.5 | Improved |
| Add MTD loss to the shrinking attack (ablation) | SiamRPN++ · LaSOT person | success performance drop: 14.5 -> 38.8 | Improved |
| Increase patch size ratio from 15% to 35% (ablation) | Universal physical attack · LaSOT | tracking performance | Improved |
| Remove the TV smoothness loss (ablation) | Universal physical attack · Physical experiments | attack effectiveness | Hurt |
| Remove the TV smoothness loss (ablation) | Universal physical attack · Digital simulations | attack performance | No clear effect |
| Use an untrained random patch instead of an optimized adversarial patch | Siamese visual trackers · LaSOT person, car, and bottle | tracking performance: Person Success: -24.7, -11.9, -2.7; Car Success: -11.9, -2.7; Bottle Success: -9.9, -16.5 | Hurt |
Closest papers by meaning
- One-Shot Adversarial Attacks on Visual Tracking With Dual Attention (CVPR 2020 · similarity 0.90)
- Physical Adversarial Textures That Fool Visual Object Tracking (ICCV 2019 · similarity 0.90)
- SPARK: Spatial-aware Online Incremental Attack Against Visual Tracking (ECCV 2020 · similarity 0.89)
- PapMOT: Exploring Adversarial Patch Attack against Multiple Object Tracking (ECCV 2024 · similarity 0.89)
- Efficient Adversarial Attacks for Visual Object Tracking (ECCV 2020 · similarity 0.88)
- Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection (NeurIPS 2024 · 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.
