Research map / Adversarial robustness and security / Attacks on text and graphs
Exploring Frequency Adversarial Attacks for Face Forgery Detection
Shuai Jia, Chao Ma, Taiping Yao, Bangjie Yin, Shouhong Ding, Xiaokang Yang
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 the frequency attack instead of spatial attacks | face forgery detectors · DFDC and FaceForensics++ | attack success rate: 41.4% on Eff_b4, 65.4% on Res50, 49.6% on Xcep, versus PGD 3.9%, 60.2%, 2.3% | Improved |
| combine spatial PGD and frequency attacks through alternating updates (ablation) | Res50-source attacks · FaceForensics++ | attack success rate: 3.9% / 60.2% / 2.3% (spatial), 14.5% / 29.3% / 15.9% (frequency), 7.7% / 61.2% / 12.1% (sum) -> 41.4% / 65.4% / 49.6% (h | Improved |
| attack low rather than middle or high frequency bands (ablation) | hybrid Res50-source attack · FaceForensics++ | attack success rate: 39.5% / 65.2% / 49.1% (low bands); 38.4% / 62.7% / 48.0% (middle); 36.2% / 61.2% / 47.1% (high) | Improved |
| attack all frequency bands instead of a single band (ablation) | hybrid Res50-source attack · FaceForensics++ | attack success rate: 41.4% / 65.4% / 49.6% (all bands), compared with low 39.5% / 65.2% / 49.1% | Improved |
| use simple summed perturbations instead of alternating hybrid updates (ablation) | spatial and frequency attack combination · FaceForensics++ | attack success rate: sum 7.7% / 61.2% / 12.1% -> hybrid 41.4% / 65.4% / 49.6% | Mixed |
| use frequency attack only rather than hybrid updates (ablation) | Res50-source attack · FaceForensics++ | attack success rate: 14.5% / 29.3% / 15.9% -> hybrid 41.4% / 65.4% / 49.6% | Hurt |
| use spatial attack only rather than hybrid updates (ablation) | Res50-source attack · FaceForensics++ | attack success rate: 3.9% / 60.2% / 2.3% -> hybrid 41.4% / 65.4% / 49.6% | Hurt |
| use the proposed method rather than FGSM or PGD | face forgery detector attacks · FaceForensics++, image quality | MSE: 0.0279 (FGSM), 0.0238 (FGD) -> 0.0027 | Improved |
| use the proposed method rather than FGSM or PGD | face forgery detector attacks · FaceForensics++, image quality | PSNR: 23.3 (FGSM), 30.4 (FGD) -> 42.7 | Improved |
| use the proposed method rather than FGSM or PGD | face forgery detector attacks · FaceForensics++, image quality | SSIM: 0.0881 (FGSM), 0.1343 (FGD) -> 0.1763 | Improved |
Closest papers by meaning
- Thinking in Frequency: Face Forgery Detection by Mining Frequency-aware Clues (ECCV 2020 · similarity 0.91)
- Generalizing Face Forgery Detection With High-Frequency Features (CVPR 2021 · similarity 0.89)
- Adaptive Face Forgery Detection in Cross Domain (ECCV 2022 · similarity 0.88)
- Frequency-Aware Discriminative Feature Learning Supervised by Single-Center Loss for Face Forgery Detection (CVPR 2021 · similarity 0.88)
- Exploring Adversarial Fake Images on Face Manifold (CVPR 2021 · similarity 0.88)
- Poisoned Forgery Face: Towards Backdoor Attacks on Face Forgery Detection (ICLR 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.
