Research map / Adversarial robustness and security / Attacks on text and graphs
Face Reconstruction Transfer Attack as Out-of-Distribution Generalization
Yoon Gyo Jung, Jaewoo Park, Xingbo Dong, Hojin Park, Andrew Beng Jin Teoh, Octavia Camps
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 |
|---|---|---|---|
| optimize 100 latents instead of a single latent (no ALSUV) (ablation) | StyleGAN2 latent search · LFW unseen encoders | SAR: 48.66 -> 92.06 | Improved |
| add latent trajectory averaging (n=100) (ablation) | ALSUV · LFW unseen encoders | SAR: 92.06 -> 93.46 | Improved |
| add unsupervised validation with pseudo target (n=100) (ablation) | ALSUV · LFW unseen encoders | SAR: 92.06 -> 94.24 | Improved |
| combine latent averaging and unsupervised validation (ablation) | ALSUV · LFW unseen encoders | SAR: 92.06 -> 95.13 | Improved |
| use 100 parallel latents x 100 steps instead of single latent with 10k steps (ablation) | latent optimization · LFW unseen encoders | SAR: 60.28 -> 95.13 | Improved |
| increase number of latents beyond 20 (ablation) | ALSUV · LFW unseen encoders | SAR: 94.48 -> 95.13 | No clear effect |
| ALSUV vs prior reconstruction methods on unseen encoders | face reconstruction transfer attack · LFW unseen encoders | SAR: 68.64 -> 86.4 | Improved |
| set k_top for validation to 10-30% of n (ablation) | ALSUV · LFW unseen encoders | SAR | Improved |
| use ALSUV instead of QEZOGE on image quality | face reconstruction · CR-FIQA | CR-FIQA: 2.262 -> 2.107 | Hurt |
| ALSUV vs DiBiGAN in white-box ResNet100 | face reconstruction · AgeDB-30 | SAR: 94.18 -> 98.33 | Improved |
Closest papers by meaning
- Recoverable Facial Identity Protection via Adaptive Makeup Transfer Adversarial Attacks (AAAI 2025 · similarity 0.88)
- Improving the Transferability of Adversarial Attacks on Face Recognition with Diverse Parameters Augmentation (CVPR 2025 · similarity 0.86)
- Transferable Facial Privacy Protection against Blind Face Restoration via Domain-Consistent Adversarial Obfuscation (ICML 2024 · similarity 0.86)
- FSFM: A Generalizable Face Security Foundation Model via Self-Supervised Facial Representation Learning (CVPR 2025 · similarity 0.84)
- Non-Adaptive Adversarial Face Generation (NeurIPS 2025 · similarity 0.84)
- From Head to Tail: Efficient Black-box Model Inversion Attack via Long-tailed Learning (CVPR 2025 · similarity 0.84)
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.
