Research map / Adversarial robustness and security / Black-box and transfer attacks
Understanding the Robustness of Skeleton-Based Action Recognition Under Adversarial Attack
He Wang, Feixiang He, Zhexi Peng, Tianjia Shao, Yong-Liang Yang, Kun Zhou, David Hogg
Check 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 derivative-matching motion dynamics and frame-wise bone-length preservation instead of joint displacement magnitude alone (ablation) | SMART skeletal-motion attacks · HDM05 and MHAD | Naturalness preference | Improved |
| Add acceleration loss to joint perturbation magnitude loss (ablation) | SMART perceptual-loss ablation · HDM05 and MHAD | Naturalness preference | Improved |
| Add bone-length loss to joint perturbation magnitude loss (ablation) | SMART perceptual-loss ablation · HDM05 and MHAD | Naturalness preference | Hurt |
| Use joint perturbation magnitude alone, without dynamics or bone-length losses (ablation) | SMART perceptual-loss ablation · HDM05 and MHAD | Naturalness preference | Hurt |
| Match 0th- and 2nd-order derivatives, omitting other derivative orders (ablation) | SMART dynamics loss · Skeletal-motion attacks | Attack and perceptual quality | Improved |
| Add 4th-order derivative matching to the 0th- and 2nd-order dynamics loss (ablation) | SMART dynamics loss · Skeletal-motion attacks | Attack and perceptual quality | Improved |
| Include consecutive derivatives such as 0th, 1st and 2nd order (ablation) | SMART dynamics loss · Skeletal-motion attacks | Optimization quality | Hurt |
| Replace SMART's perceptual loss with magnitude-only perturbation constraints (ablation) | SMART attack · HDM05 | Indistinguishability | Hurt |
| Use AB5 rather than AB for transfer attacks | SMART black-box attack · NTU60 | Black-box success rate: 90.6(90.99); 7.24(7.63); 98.37(98.46); 98.10(98.96); 10.90(12.97); 91.17(91.99) | Improved |
| Use SMART instead of IAA | HDM05 AB attack · HDM05 | Indistinguishability: 42.22% vs 36.67%; 90.00% vs 87.5%; 80.83% vs 35.33% | Improved |
| Use SMART instead of CIASA | HDM05 AB attack · HDM05 | Indistinguishability: 42.22% vs 32.22%; 90.00% vs 90.00%; 80.83% vs 49.33% | Mixed |
| Use a dynamics-aware perceptual loss that permits larger joint deviations (ablation) | SMART skeletal-motion attack · HDM05 | Naturalness preference | Improved |
Closest papers by meaning
- Adversarial Bone Length Attack on Action Recognition (AAAI 2022 · similarity 0.92)
- BASAR:Black-Box Attack on Skeletal Action Recognition (CVPR 2021 · similarity 0.91)
- Hard No-Box Adversarial Attack on Skeleton-Based Human Action Recognition with Skeleton-Motion-Informed Gradient (ICCV 2023 · similarity 0.91)
- TASAR: Transfer-based Attack on Skeletal Action Recognition (ICLR 2025 · similarity 0.89)
- MaskCLR: Attention-Guided Contrastive Learning for Robust Action Representation Learning (CVPR 2024 · similarity 0.85)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.84)
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.
