Research map / Federated learning and privacy / Privacy attacks and secure learning
Faceless Person Recognition: Privacy Implications in Social Media
Seong Joon Oh, Rodrigo Benenson, Mario Fritz, Bernt Schiele
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 |
|---|---|---|---|
| fine-tune pair matcher on tagged samples split0 (ablation) | Siamese pair matching, visible · PIPA pairs within events | EER (%): 79.1 -> 92.7 | Improved |
| fine-tune pair matcher on tagged samples split0 (ablation) | Siamese pair matching, visible · PIPA pairs across events | EER (%): 74.5 -> 81.4 | Improved |
| add pairwise terms with negative edge pruning (ablation) | Faceless recognition CRF · PIPA val S2 black, across events | accuracy (%): 14.7 -> 39.2 | Improved |
| add pairwise terms with negative edge pruning (ablation) | Faceless recognition CRF · PIPA val S2 black, within events | accuracy (%): 80.9 -> 81.9 | Improved |
| add pairwise terms without negative edge pruning (ablation) | Faceless recognition CRF · PIPA val S2 black, within events | accuracy (%): 81.9 -> 47.3 | Hurt |
| add negative edge pruning to pairwise terms (ablation) | Faceless recognition CRF · PIPA val S2 black, across events | accuracy (%): 29.4 -> 39.2 | Improved |
| adapt convnet to blurred heads (ablation) | Unary person recogniser · PIPA val blur, across events | accuracy (%): 23.5 -> 28.8 | Improved |
| adapt convnet to blurred heads (ablation) | Unary person recogniser · PIPA val blur, within events | accuracy (%): 84.3 -> 86.7 | Improved |
| adapt convnet to black fill-in (ablation) | Unary person recogniser · PIPA val black, within events | accuracy (%): 80.1 -> 80.9 | No clear effect |
| use white instead of black fill-in obfuscation | Unary person recogniser · PIPA val within events adapted | accuracy (%): 80.9 -> 79.6 | No clear effect |
| increase tag rate from 1.25 to 10 instances per identity | Unary+Pairwise · PIPA test visible, within events | accuracy (%): 56.8 -> 86.4 | Improved |
| add pairwise terms (ablation) | Faceless recognition CRF · PIPA test S3 black, across events | accuracy (%): 19.6 -> 23.9 | Improved |
Closest papers by meaning
- PerceptAnon: Exploring the Human Perception of Image Anonymization Beyond Pseudonymization for GDPR (ICML 2024 · similarity 0.86)
- Person De-reidentification: A Variation-guided Identity Shift Modeling (CVPR 2025 · similarity 0.85)
- Validating Privacy-Preserving Face Recognition under a Minimum Assumption (CVPR 2024 · similarity 0.84)
- Disguise without Disruption: Utility-Preserving Face De-identification (AAAI 2024 · similarity 0.84)
- MultiPriv: Benchmarking Individual-Level Privacy Reasoning in Vision-Language Models (ICML 2026 · similarity 0.84)
- Privacy-Preserving Face Recognition Using Trainable Feature Subtraction (CVPR 2024 · 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.
