Research map / Adversarial robustness and security / Backdoor and poisoning
Clean-Label Backdoor Attacks on Video Recognition Models
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace the transferred clean-label image attack with a universal adversarial trigger and targeted perturbation | I3D · UCF-101 | attack success rate (%): 1.1 -> 82.2 | Helped |
| Add a universal adversarial trigger rather than a fixed or randomly sampled trigger (ablation) | I3D with targeted perturbation · UCF-101 | attack success rate (%): 0.0 -> 82.2; 0.0 -> 82.2 | Helped |
| Add a universal adversarial trigger rather than a fixed or randomly sampled trigger (ablation) | I3D with targeted perturbation · HMDB-51 | attack success rate (%): 1.4 -> 81.0; 4.1 -> 81.0 | Helped |
| Add targeted adversarial perturbation to the universal adversarial trigger (ablation) | I3D · UCF-101 | attack success rate (%): 61.1 -> 82.2 | Helped |
| Add uniform adversarial perturbation to the universal adversarial trigger (ablation) | I3D · UCF-101 | attack success rate (%): 61.1 -> 76.9 | Helped |
| Add targeted adversarial perturbation to the universal adversarial trigger (ablation) | I3D · HMDB-51 | attack success rate (%): 69.2 -> 81.0 | Helped |
| Add uniform adversarial perturbation to the universal adversarial trigger (ablation) | I3D · HMDB-51 | attack success rate (%): 69.2 -> 81.7 | Helped |
| Increase trigger size from 16 to 28 (ablation) | Proposed video backdoor attack · UCF-101 | attack success rate (%): 61.2 -> 93.7 | Helped |
| Use optical flow alongside RGB input (ablation) | I3D two-stream network · UCF-101 | attack success rate (%): 68.5 -> 54.7 | Hurt |
| Apply the attack to optical-flow input rather than RGB input (ablation) | I3D · UCF-101 | attack success rate (%): 68.5 -> 15.2 | Hurt |
| Train with data augmentation (ablation) | Proposed video backdoor attack · UCF-101 | attack success rate (%): 56.3% | Hurt |
| Increase poisoning percentage beyond the reported 20%–70% range (ablation) | Proposed video backdoor attack · UCF-101 | attack success rate (%): nearly 0% | Hurt |
Closest papers by meaning
- Clean-image Backdoor: Attacking Multi-label Models with Poisoned Labels Only (ICLR 2023 · similarity 0.93)
- Hidden Trigger Backdoor Attacks (AAAI 2020 · similarity 0.92)
- Temporal-Distributed Backdoor Attack against Video Based Action Recognition (AAAI 2024 · similarity 0.91)
- Wicked Oddities: Selectively Poisoning for Effective Clean-Label Backdoor Attacks (ICLR 2025 · similarity 0.91)
- A Theoretical Analysis of Backdoor Poisoning Attacks in Convolutional Neural Networks (ICML 2024 · similarity 0.90)
- Adversarial-Inspired Backdoor Defense via Bridging Backdoor and Adversarial Attacks (AAAI 2025 · similarity 0.89)
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.