Research map / Adversarial robustness and security / Attacks on text and graphs
Motion-Excited Sampler: Video Adversarial Attack with Sparked Prior
Hu Zhang, Linchao Zhu, Yi Zhu, Yi Yang
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 motion-excited sampler (motion maps) instead of per-frame random noise (ablation) | ME-Sampler · UCF-101 vs I3D | ANQ: 15,798 -> 6,876 | Improved |
| add motion maps instead of one noise replicated to all frames (ablation) | ME-Sampler · UCF-101 vs I3D | ANQ: 22,908 -> 6,876 | Improved |
| add motion maps instead of one noise replicated to all frames (ablation) | ME-Sampler · UCF-101 vs I3D | success rate: 93.33 -> 100.0 | Improved |
| use true motion map instead of uniformly sampled values with same binary mask (ablation) | ME-Sampler · UCF-101 vs TSN2D | ANQ and success rate: 9,319 queries saved and 16.77% higher success rate | Improved |
| use sequenced-value map instead of uniform-random map (ablation) | ME-Sampler · Kinetics-400 vs I3D | ANQ and success rate: saves more than 1,000 queries but gets 2% higher success rate | Improved |
| use true motion map instead of sequenced-value map (ablation) | ME-Sampler · UCF-101 vs I3D | ANQ: 11,471 -> 6,876 | Improved |
| use original motion vector instead of randomized motion map (ablation) | ME-Sampler · UCF-101 vs I3D | ANQ: 6,970 -> 6,876 | No clear effect |
| use motion vector instead of optical flow | ME-Sampler · UCF-101 vs TSN2D | ANQ: 6,598 -> 5,353 | Improved |
| use motion vector instead of optical flow | ME-Sampler · HMDB-51 vs I3D | ANQ: 3,306 -> 3,915 | Hurt |
| use ME-Sampler (MV) instead of V-BAD | ME-Sampler · SthSth-V2 vs I3D | ANQ and success rate: 7,239 -> 592 queries, 97.70 -> 100.0 SR | Improved |
| use ME-Sampler (MV) instead of E-Bandits | ME-Sampler · UCF-101 vs TSN2D | ANQ and success rate: 24,890 -> 5,353 queries, 66.33 -> 99.00 SR | Improved |
| use ME-Sampler instead of V-BAD in targeted attack | ME-Sampler · HMDB-51 targeted vs TSN2D | success rate: 64.86 -> 92.16 | Improved |
Closest papers by meaning
- Generating Videos with Scene Dynamics (NeurIPS 2016 · similarity 0.86)
- Adversarial Framework for Unsupervised Learning of Motion Dynamics in Videos (IJCV 2019 · similarity 0.86)
- Boosting Adversarial Transferability using Dynamic Cues (ICLR 2023 · similarity 0.86)
- Hierarchical Video Generation From Orthogonal Information: Optical Flow and Texture (AAAI 2018 · similarity 0.86)
- FrameShield: Adversarially Robust Video Anomaly Detection (NeurIPS 2025 · similarity 0.84)
- OSV: One Step is Enough for High-Quality Image to Video Generation (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.
