Research map / Adversarial robustness and security / Black-box and transfer attacks
Global-Local Characteristic Excited Cross-Modal Attacks from Images to Videos
Ruikui Wang, Yuanfang Guo, Yunhong 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 |
|---|---|---|---|
| Add global inter-frame interaction enhancement (GIE) (ablation) | I2V-style cross-modal video attacks · UCF-101 and Kinetics-400 | Average ASR (%): improved by 2.2% ∼8.5% and 1.5% ∼9.6% | Improved |
| Add local correlation disturbance (LCD) (ablation) | I2V-style cross-modal video attacks · UCF-101 and Kinetics-400 | Average ASR (%): 6.1% | Mixed |
| Combine GIE and LCD (ablation) | I2V baseline · UCF-101 and Kinetics-400 | Average ASR (%): 6.5% ∼10.1% and 3.8% ∼13.6% | Improved |
| Use GIE with n1=2 and n2=3 rather than the baseline (ablation) | VGG-16 substitute-model attack · UCF-101 and Kinetics-400 | Average ASR (%): 36.63 -> 44.06; 56.42 -> 66.04 | Improved |
| Increase the number of GIE interaction combinations from n2=2 to n2=3 with n1=2 (ablation) | VGG-16 substitute-model attack · UCF-101 and Kinetics-400 | Average ASR (%): 45.05 -> 44.06; 65.67 -> 66.04 | Mixed |
| Increase the number of interacted frames from n1=2 to n1=3 with n2=2 (ablation) | VGG-16 substitute-model attack · UCF-101 and Kinetics-400 | Average ASR (%): 45.05 -> 42.90; 65.67 -> 64.46 | Hurt |
| Add LCD with λ=0.1 (ablation) | VGG-16 substitute-model baseline · UCF-101 and Kinetics-400 | Average ASR (%): 36.63 -> 35.14; 56.42 -> 56.33 | Hurt |
| Add LCD with λ=0.01 (ablation) | VGG-16 substitute-model baseline · UCF-101 and Kinetics-400 | Average ASR (%): 36.63 -> 38.61; 56.42 -> 58.75 | Improved |
| Add LCD with λ=0.001 (ablation) | VGG-16 substitute-model baseline · UCF-101 and Kinetics-400 | Average ASR (%): 36.63 -> 37.29; 56.42 -> 56.33 | Mixed |
| Add GIE and LCD to I2V (ablation) | Cross-modal attacks using DenseNet-121 · Kinetics-400 | Average ASR (%): I2V: 81.69; I2V+LCD: 82.00; GIE: 82.63; GIE+LCD: 84.00 | Improved |
| Use GIE+LCD with DenseNet-121+ResNet-101 instead of DenseNet-121 | GIE+LCD · Kinetics-400 | Average ASR (%): 84.00 -> 85.50 | Improved |
Closest papers by meaning
- Cross-Modal Transferable Adversarial Attacks From Images to Videos (CVPR 2022 · similarity 0.94)
- Adaptive Cross-Modal Transferable Adversarial Attacks From Images to Videos (TPAMI 2023 · similarity 0.94)
- Transferability of Adversarial Attacks in Video-based MLLMs: A Cross-modal Image-to-Video Approach (AAAI 2026 · similarity 0.89)
- Everywhere Attack: Attacking Locally and Globally to Boost Targeted Transferability (AAAI 2025 · similarity 0.87)
- Mutual-Modality Adversarial Attack with Semantic Perturbation (AAAI 2024 · similarity 0.86)
- I2VGuard: Safeguarding Images against Misuse in Diffusion-based Image-to-Video Models (CVPR 2025 · similarity 0.86)
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.
