Research map / Video / Action recognition
Learning to Anonymize Faces for Privacy Preserving Action Detection
Zhongzheng Ren, Yong Jae Lee, Michael S. Ryoo
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 |
|---|---|---|---|
| use learned adversarial face modifier instead of masking | Face anonymization · DALY | action detection mAP: 38.86 -> 52.32 | Improved |
| use learned modifier instead of 24x24 blur | Face anonymization · DALY | action detection mAP: 51.97 -> 52.32 | Improved |
| use learned modifier instead of 8x8 blur | Face anonymization · DALY | action detection mAP: 42.39 -> 52.32 | Improved |
| apply any anonymization versus none | Faster R-CNN action detector · DALY | action detection mAP: 56.85 -> 52.32 | Hurt |
| learned modifier vs Gaussian noise 0.5 | Face anonymization · DALY | action detection mAP: 43.81 -> 52.32 | Improved |
| use learned modifier instead of masking on drinking | Face anonymization · DALY drinking | AP: 6.59 -> 22.93 | Improved |
| use learned modifier instead of masking on playing harmonica | Face anonymization · DALY harmonica | AP: 21.32 -> 55.32 | Improved |
| use learned modifier on face-irrelevant action (cleaning floor) instead of blur | Face anonymization · DALY cleaning floor | AP: 79.93 -> 77.12 | No clear effect |
| optimize modifier with gradient ascent on identity loss instead of random-label target (ablation) | Face modifier · faces | visual quality | Improved |
| train face classifier on both original and modified faces (ablation) | Face classifier · LFW | verification accuracy: 94.75 -> 66.95 (original vs modified faces) | Improved |
| modified faces shown to humans | Learned anonymizer · User study | same-person accuracy: 53.3% | Improved |
Closest papers by meaning
- Privacy Beyond Pixels: Latent Anonymization for Privacy-Preserving Video Understanding (ICLR 2026 · similarity 0.92)
- Privacy-Preserving Deep Action Recognition: An Adversarial Learning Framework and A New Dataset (TPAMI 2020 · similarity 0.92)
- ReGenHuman: Re-Generating Human Appearances for Realistic Full-Body Video Anonymization (NeurIPS 2026 · similarity 0.86)
- Balancing Privacy and Performance: A Many-in-One Approach for Image Anonymization (AAAI 2025 · similarity 0.86)
- Privacy-Preserving Optics for Enhancing Protection in Face De-Identification (CVPR 2024 · similarity 0.85)
- PerceptAnon: Exploring the Human Perception of Image Anonymization Beyond Pseudonymization for GDPR (ICML 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.
