Research map / Adversarial robustness and security / Attacks on text and graphs
Breaking Temporal Consistency: Generating Video Universal Adversarial Perturbations Using Image Models
Hee-Seon Kim, Minji Son, Minbeom Kim, Myung-Joon Kwon, Changick Kim
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 Temporal Similarity Loss (BTC) to image-model UAP optimization, versus I2V-UAP | Res-101 UAP, ImageNet · Kinetics-400, avg over 6 video models | ASR (%): 65.90 -> 79.31 | Improved |
| add Temporal Similarity Loss (BTC) versus I2V-UAP with UCF-101 source data | Res-101 UAP · UCF-101 unseen group, avg over 6 video models | ASR (%): 28.56 -> 51.12 | Improved |
| generate UAP from image model and image data (BTC) instead of video models and data | BTC-UAP · Kinetics-400, avg ASR | ASR (%): 70.88 (TT-UAP SF-101) -> 79.31 | Improved |
| use BTC-UAP with VGG16 source model instead of I2V-UAP | VGG16 UAP · Kinetics-400, avg | ASR (%): 52.06 -> 73.41 | Improved |
| use BTC-UAP with SqueezeNet source model instead of the best baseline | SqueezeNet UAP · Kinetics-400, avg | ASR (%): 59.99 -> 64.83 | Improved |
| add Feature Diversity with K=4 random noises (ablation) | BTC-UAP adversarial loss · UCF-101 | ASR (%) | Improved |
| use neighbor distance set J={-2,-1,1,2} instead of smaller sets (ablation) | BTC-UAP temporal loss · UCF-101 | ASR (%) | Improved |
| use temporal loss J={-1,1} with K=4 instead of K=6 at same compute (ablation) | BTC-UAP · UCF-101 | ASR (%) | Improved |
| reduce number of UAP frames N from 32 to 2 (repeated) (ablation) | BTC-UAP · UCF-101 | ASR (%) | No clear effect |
| shift the UAP along the temporal axis by 1 to 8 frames (ablation) | BTC-UAP · UCF-101 | ASR (%) | No clear effect |
| repeat 32-frame UAP to cover 64-frame Kinetics videos without re-optimizing | BTC-UAP · Kinetics-400 | ASR (%) | No clear effect |
Closest papers by meaning
- Data-free Universal Adversarial Perturbation with Pseudo-semantic Prior (CVPR 2025 · similarity 0.86)
- Data-Free Universal Attack by Exploiting the Intrinsic Vulnerability of Deep Models (AAAI 2025 · similarity 0.86)
- Vanish into Thin Air: Cross-prompt Universal Adversarial Attacks for SAM2 (NeurIPS 2025 · similarity 0.85)
- Generate Universal Adversarial Perturbations for Few-Shot Learning (NeurIPS 2024 · similarity 0.85)
- Improving Generalization of Universal Adversarial Perturbation via Dynamic Maximin Optimization (AAAI 2025 · similarity 0.85)
- FrameShield: Adversarially Robust Video Anomaly Detection (NeurIPS 2025 · 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.
