Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial Bone Length Attack on Action Recognition
Nariki Tanaka, Hiroshi Kera, Kazuhiko Kawamoto
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 |
|---|---|---|---|
| Train with data augmentation (ablation) | ST-GCN on HDM05 · HDM05 | clean accuracy: 86.1% -> 89.8% | Improved |
| Train with data augmentation (ablation) | SGN on HDM05 · HDM05 | clean accuracy: 95.3% -> 96.1% | Improved |
| Train with data augmentation (ablation) | ST-GCN on HDM05 · HDM05 bone-length attack | attack success rate | Hurt |
| Train with data augmentation (ablation) | SGN on HDM05 · HDM05 bone-length attack | attack success rate | Hurt |
| Adversarially train with bone-length attacks at ε=0.1 (ablation) | ST-GCN on HDM05 · HDM05 | clean accuracy: 86.1% -> 89.8% | Improved |
| Adversarially train with bone-length attacks at ε=0.1 (ablation) | SGN on HDM05 · HDM05 | clean accuracy: 96.1% -> 95.5% | Hurt |
| Adversarially train with bone-length attacks at ε=0.1 (ablation) | ST-GCN on HDM05 · HDM05 bone-length attack | attack success rate | Hurt |
| Adversarially train with bone-length attacks at ε=0.1 (ablation) | SGN on HDM05 · HDM05 bone-length attack | attack success rate | Hurt |
| Restrict perturbations to Part 6 (legs) rather than other individual parts (ablation) | Bone length attack on ST-GCN · HDM05 | attack success rate: 34.1% vs 0.00% to 12.6% at ε=0.1 | Improved |
| Perturb Parts 3, 4, and 5 together instead of Part 6 (ablation) | Bone length attack on ST-GCN · HDM05, ε=0.3 | attack success rate: 75.8% -> 59.6% | Hurt |
| Continue the attack for the full run rather than stop at first success (ablation) | Bone length attack on ST-GCN · HDM05 adversarial examples | average confidence: 51.3% -> 50.4% | No clear effect |
| Perturb bones that are longer and closer to the root joint | Bone length attack · HDM05 | attack success rate | Improved |
Closest papers by meaning
- Understanding the Robustness of Skeleton-Based Action Recognition Under Adversarial Attack (CVPR 2021 · similarity 0.92)
- 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 0.88)
- Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton Action Recognition (CVPR 2025 · similarity 0.84)
- MaskCLR: Attention-Guided Contrastive Learning for Robust Action Representation Learning (CVPR 2024 · similarity 0.83)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.83)
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.
