Research map / Adversarial robustness and security / Backdoor and poisoning
Towards Physical World Backdoor Attacks against Skeleton Action Recognition
Qichen Zheng, Yi Yu, SIYUAN YANG, Jun Liu, Kwok-Yan Lam, Alex Kot
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 |
|---|---|---|---|
| increase poisoning rate from 0.1% to 5% in P-PSBA (nodding trigger) (ablation) | P-PSBA · NTU RGB+D (Hyperformer) | ASR (%): 14.72 -> 98.55 | Improved |
| increase poisoning rate in P-PSBA (ablation) | P-PSBA · NTU RGB+D (Hyperformer) | clean accuracy (%): 90.67 -> 90.53 (5% nodding) | No clear effect |
| increase clean-label poisoning ratio from 30% to 90% of target class data (ablation) | C-PSBA · NTU RGB+D (Hyperformer, crossing hands) | ASR (%): 64.34 -> 79.06 | Improved |
| use clean-label instead of poison-label attack | PSBA · NTU RGB+D, NTU RGB+D 120, PKU-MMD | ASR (%) | Hurt |
| add adversarial-perturbation trigger-enhancing strategy to clean-label attack (ablation) | C-PSBA · NTU RGB+D | ASR (%) | Improved |
| apply CLP pruning defense against P-PSBA | P-PSBA (bending trigger) · NTU RGB+D | ASR (%): 88.37 -> 88.19 (Hyperformer, 1%) | No clear effect |
| apply D-BR fine-tuning defense against P-PSBA | P-PSBA (bending trigger) · NTU RGB+D | ASR (%): 92.88 -> 92.50 (Hyperformer, 2%) | No clear effect |
| apply STRIP input-filtering defense against P-PSBA | P-PSBA (2% poison) · NTU RGB+D (INFO-GCN) | entropy distribution separation | No clear effect |
| increase poisoning rate from 1% to 2% on real-world captured data (ablation) | P-PSBA · Real-world Kinect data, NTU RGB+D (nodding, Hyperformer) | physical ASR (%): 41.87 -> 71.40 | Improved |
| choose nodding as trigger instead of bending sideways at low poison rate (ablation) | P-PSBA · NTU RGB+D (Hyperformer, 0.2%) | ASR (%): 54.33 -> 23.98 | Hurt |
| raise poisoning ratio, stealthiness distribution distance grows (ablation) | P-PSBA · NTU RGB+D 120 | EMD (10^-3): 5 -> 139 (bending, 0.1% to 5%) | Hurt |
Closest papers by meaning
- TASAR: Transfer-based Attack on Skeletal Action Recognition (ICLR 2025 · similarity 0.89)
- Adversarial Bone Length Attack on Action Recognition (AAAI 2022 · similarity 0.87)
- Understanding the Robustness of Skeleton-Based Action Recognition Under Adversarial Attack (CVPR 2021 · similarity 0.87)
- BASAR:Black-Box Attack on Skeletal Action Recognition (CVPR 2021 · similarity 0.87)
- Navigating Open Set Scenarios for Skeleton-Based Action Recognition (AAAI 2024 · similarity 0.83)
- Skeleton-based Action Recognition with Non-linear Dependency Modeling and Hilbert-Schmidt Independence Criterion (AAAI 2025 · similarity 0.82)
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.
