Research map / Adversarial robustness and security / Attacks on text and graphs
Over-the-Air Adversarial Flickering Attacks Against Video Recognition Networks
Roi Pony, Itay Naeh, Shie Mannor
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 an optimized universal flickering perturbation instead of random MinMax flickering | Video action recognition models · Kinetics-400 evaluation, $\ell_{\infty}[%]=20$ | fooling ratio: 48.0$\pm$ 4.5% -> 93.0% | Improved |
| Use an optimized universal flickering perturbation instead of random MinMax flickering | Video action recognition models · Kinetics-400 evaluation, $\ell_{\infty}[%]=20$ | fooling ratio: 42.0$\pm$ 3.0% -> 79.0% | Improved |
| Use an optimized universal flickering perturbation instead of random MinMax flickering | Video action recognition models · Kinetics-400 evaluation, $\ell_{\infty}[%]=20$ | fooling ratio: 54.6$\pm$ 11.0% -> 90.3% | Improved |
| Use an optimized universal flickering perturbation instead of random MinMax flickering | Video action recognition models · Kinetics-400 evaluation, $\ell_{\infty}[%]=20$ | fooling ratio: 44.0$\pm$ 5.0% -> 77.1% | Improved |
| Add a random temporal shift during training and evaluation (ablation) | Universal flickering attack on I3D · Kinetics-400 test | fooling ratio: 93.0 -> 83.1 | Hurt |
| Add roughness regularization rather than minimizing thickness alone (ablation) | Single-video flickering attack on I3D · Kinetics-400 test, single video example | roughness: 4.84\% -> 2.20\% | Improved |
| Add roughness regularization rather than minimizing thickness alone (ablation) | Single-video flickering attack on I3D · Kinetics-400 test, single video example | thickness: 2.97\% -> 7.45\% | Hurt |
| Use one universal perturbation across videos instead of video-specific perturbations | Flickering attack on I3D · Kinetics-400 | fooling ratio: 100 -> 93.0 | Mixed |
| Use a class-specific generalized perturbation instead of a video-specific perturbation | Flickering attack on I3D · Kinetics-400 | fooling ratio: 100 -> 90.2$\pm$ 11.72 | Mixed |
Closest papers by meaning
- Sparse Adversarial Perturbations for Videos (AAAI 2019 · similarity 0.88)
- Breaking Temporal Consistency: Generating Video Universal Adversarial Perturbations Using Image Models (ICCV 2023 · similarity 0.86)
- AdvIT: Adversarial Frames Identifier Based on Temporal Consistency in Videos (ICCV 2019 · similarity 0.85)
- Boosting the Transferability of Video Adversarial Examples via Temporal Translation (AAAI 2022 · similarity 0.85)
- Merry Go Round: Rotate a Frame and Fool a DNN (CVPR 2022 · similarity 0.85)
- FrameShield: Adversarially Robust Video Anomaly Detection (NeurIPS 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.
