Research map / Adversarial robustness and security / Attacks on text and graphs
AdvIT: Adversarial Frames Identifier Based on Temporal Consistency in Videos
Chaowei Xiao, Ruizhi Deng, Bo Li, Taesung Lee, Benjamin Edwards, Jinfeng Yi, Dawn Song, Mingyan Liu, Ian Molloy
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 |
|---|---|---|---|
| use optical-flow pseudo frames with randomness instead of JPEG compression | AdvIT · Object detection (DAG Person) | detection AUC: 60% -> 98% | Improved |
| use optical-flow pseudo frames instead of replacing the frame with previous frames | AdvIT · Segmentation (Houdini Stripe) | detection AUC: 50% -> 100% | Improved |
| use flow-warped pseudo frames instead of plain previous-frame replacement | AdvIT · Object detection (DAG Person) | detection AUC: 50% -> 98% | Improved |
| aggregate more previous frames (k from 1 to 5) | AdvIT · Object detection (DAG Person) | detection AUC: 98% -> 100% | Improved |
| aggregate more previous frames (k from 1 to 5) | AdvIT · Sparse attack, action recognition | detection AUC: 95% -> 97% | Improved |
| aggregate more previous frames under adaptive attack with adversarial previous frames | AdvIT · Adaptive attack, pose Shuffle | detection AUC: 94% -> 100% | Improved |
| aggregate more previous frames under adaptive attack | AdvIT · Adaptive attack, segmentation ICCV | detection AUC: 95% -> 100% | Improved |
| adaptive attacker using expectation over transformation versus non-adaptive attack | AdvIT · Segmentation, pose, detection | detection AUC: above 95% | No clear effect |
| use a non-differentiable flow estimator against a differentiable-flow attacker | AdvIT · Transferability (Table 3) | detection AUC: 100% | Improved |
| increase perturbation magnitude to 2, 16, 32 pixels | AdvIT · Perturbation strength | detection rate: decrease a bit | Hurt |
| prior frames adversarial versus benign | AdvIT · Segmentation, pose, detection | detection AUC: almost 100% | No clear effect |
| add detection with k pseudo-frame inferences | AdvIT · Segmentation run time | seconds: 2.58 -> 2.98 | Hurt |
Closest papers by meaning
- Breaking Temporal Consistency: Generating Video Universal Adversarial Perturbations Using Image Models (ICCV 2023 · similarity 0.88)
- Merry Go Round: Rotate a Frame and Fool a DNN (CVPR 2022 · similarity 0.86)
- Attacking Video Recognition Models with Bullet-Screen Comments (AAAI 2022 · similarity 0.86)
- FrameShield: Adversarially Robust Video Anomaly Detection (NeurIPS 2025 · similarity 0.86)
- Connecting the Dots: Detecting Adversarial Perturbations Using Context Inconsistency (ECCV 2020 · similarity 0.86)
- Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection (NeurIPS 2024 · 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.
