Research map / Adversarial robustness and security / Certified and empirical robustness
Proactive Defense Benchmark against Deepfake Generation
Joonhyuk Baek, Wonjune Seo, Jae-yun Kim, Saerom Park, Hoki Kim
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 robustness constraints to the defense (Disrupting, DF-RAP versus PGD) | proactive defenses · CelebA-HQ, SimSwap | ROB: 0.7269 -> 0.9882 | Improved |
| add robustness constraints to the defense | proactive defenses · white-box disruption (CelebA-HQ) | MSE: 1.3715 -> 0.3003 (StarGAN) | Hurt |
| use optimization-based pixel attacks on diffusion generators instead of GANs | PGD · CelebA-HQ, DiffSwap | CIDR: 0.4407 (SimSwap) -> 0.0363 (DiffSwap) | Hurt |
| train perturbation on a single source generator and transfer to unseen generators | single-source defenses · cross-generator transfer | disruption | Hurt |
| use ensemble-based defense CMUA instead of single-source PGD | PGD · black-box transfer, StyleCLIP | CIDR: +0.0513 | Improved |
| use ensemble-based defense CMUA instead of single-source PGD | PGD · white-box DiffAE | MSE: -0.0975 | Hurt |
| use LEAT ensemble instead of single-source Latent Attack | Latent Attack · black-box transfer, StyleCLIP | CIDR: +0.0231 | Improved |
| use calibrated identity ratio (CIDR) instead of raw ID loss | identity evaluation · deepfake defense benchmark | identity disruption metric | Improved |
| evaluate with identity metric instead of fidelity metric | defense ranking · DiffAE, NullSwap | CIDR vs MSE: CIDR 0.5028 vs MSE 0.0053 | Mixed |
| extend face-swap defense SCOL to attribute manipulation | SCOL · CelebA-HQ, DiffAE | CIDR: 0.3769 | Improved |
| apply Anti-Forgery to generators other than StarGAN | Anti-Forgery · CelebA-HQ, DiffAE and SimSwap | disruption: MSE 1.6180 (StarGAN) -> 0.0020 (DiffAE) | Hurt |
Closest papers by meaning
- D^3: Scaling Up Deepfake Detection by Learning from Discrepancy (CVPR 2025 · similarity 0.83)
- Where the Devil Hides: Deepfake Detectors Can No Longer Be Trusted (CVPR 2025 · similarity 0.83)
- SWE-ABS: Adversarial Benchmark Strengthening Exposes Inflated Success Rates on Test-based Benchmark (ICML 2026 · similarity 0.83)
- PubDef: Defending Against Transfer Attacks From Public Models (ICLR 2024 · similarity 0.82)
- TriDF: Evaluating Perception, Detection, and Hallucination for Interpretable DeepFake Detection (CVPR 2026 · similarity 0.82)
- MOSAIC-Bench: Measuring Compositional Vulnerability Induction in Coding Agents (NeurIPS 2026 · similarity 0.82)
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.
