Research map / Federated learning and privacy / Privacy attacks and secure learning
OPOM: Customized Invisible Cloak Towards Face Privacy Protection
Yaoyao Zhong, Weihong Deng
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 |
|---|---|---|---|
| model identity subspace as convex hull instead of a single point (ablation) | FI-UAP · Privacy-Commons, Softmax source, ArcFace target (Top-1) | Top-1 protection success rate (%): 72.3 -> 78.0 | Improved |
| use convex hull instead of affine hull for the identity subspace (ablation) | OPOM · Privacy-Commons, Softmax source, ArcFace target (Top-1) | Top-1 protection success rate (%): 73.0 -> 78.0 | Improved |
| use convex hull instead of class center (ablation) | OPOM · Privacy-Commons, Softmax source, ArcFace target (Top-1) | Top-1 protection success rate (%): 76.9 -> 78.0 | Improved |
| add intra-class interactions (cross-image gradients) to FI-UAP (ablation) | FI-UAP · Privacy-Commons, Softmax source, ArcFace target (Top-1) | Top-1 protection success rate (%): 72.3 -> 76.9 | Improved |
| use class center versus FI-UAP+ intra-class formulation (ablation) | OPOM-ClassCenter · Privacy-Commons, Softmax source, ArcFace target (Top-1) | Top-1 protection success rate (%): 76.9 -> 76.9 | No clear effect |
| aggregate all training images into one universal mask instead of per-identity mask (ablation) | FI-UAP · Privacy-Commons, Softmax source, ArcFace target (Top-1) | Top-1 protection success rate (%): 72.3 -> 53.4 | Hurt |
| use convex hull instead of single-point FI-UAP (ablation) | OPOM · Privacy-Commons, CosFace source, MobileNet target (Top-1) | Top-1 protection success rate (%): 55.6 -> 62.3 | Improved |
| use convex hull instead of affine hull (ablation) | OPOM · Privacy-Commons, CosFace source, MobileNet target (Top-1) | Top-1 protection success rate (%): 58.6 -> 62.3 | Improved |
| use convex hull instead of single-point FI-UAP (ablation) | OPOM · Privacy-Celebrities, Softmax source, ArcFace target (Top-1) | Top-1 protection success rate (%): 56.5 -> 63.8 | Improved |
| use convex hull instead of single-point FI-UAP (ablation) | OPOM · Privacy-Celebrities, ArcFace source, SENet target (Top-1) | Top-1 protection success rate (%): 54.3 -> 62.3 | Improved |
| add momentum boosting and DFANet transferability methods | OPOM-ConvexHull · Privacy-Commons, Softmax source, ArcFace target (Top-1) | Top-1 protection success rate (%): 78.0 -> 81.2 | Improved |
Closest papers by meaning
- Privacy-Preserving Face Recognition Using Trainable Feature Subtraction (CVPR 2024 · similarity 0.84)
- Disguise without Disruption: Utility-Preserving Face De-identification (AAAI 2024 · similarity 0.84)
- Enhancing Facial Privacy Protection via Weakening Diffusion Purification (CVPR 2025 · similarity 0.83)
- Privacy-Preserving Optics for Enhancing Protection in Face De-Identification (CVPR 2024 · similarity 0.83)
- Asymmetric Invertible Threat: Learning Reversible Privacy Defense for Face Recognition (NeurIPS 2026 · similarity 0.83)
- Transferable Facial Privacy Protection against Blind Face Restoration via Domain-Consistent Adversarial Obfuscation (ICML 2024 · 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.
