Research map / Federated learning and privacy / Privacy attacks and secure learning
Learning Phase Mask for Privacy-Preserving Passive Depth Estimation
Zaid Tasneem, Giovanni Milione, Yi-Hsuan Tsai, Xiang Yu, Ashok Veeraraghavan, Manmohan Chandraker, Francesco Pittaluga
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 heavy defocus (no phase mask) to the sensor | all-in-focus sensor · VGGFace2 face identification | face identification accuracy (%): 80.1% -> 21.7% | Improved |
| add privacy adversary to phase mask optimization (sensor D vs depth-only sensor C) (ablation) | depth-optimized phase mask · VGGFace2 face identification | face identification accuracy (%): 40% -> 20.5% | Improved |
| use a random phase mask instead of no mask (ablation) | sensor A · depth estimation and face identification | privacy-utility trade-off | Hurt |
| use a larger undiffracted PSF (focal distance u=0.1m vs 0.17 and 1.0m) (ablation) | learned phase mask sensor · privacy-utility trade-off | privacy leakage | Improved |
| reduce sensor resolution from 64x64 to 16x16 (ablation) | learned phase mask sensor · VGGFace2 face identification | face identification accuracy | Improved |
| reduce sensor resolution from 64x64 to 16x16 (ablation) | learned phase mask sensor · NYUv2 depth | depth estimation quality | No clear effect |
| replace all-in-focus sensor with learned sensor H | all-in-focus sensor · VGGFace2 face identification | face identification accuracy (%): 80.1% -> 5.3% | Improved |
| replace conventional sensor with learned sensor for face verification attack | conventional sensor · LFW face verification | AUC: 0.99 -> 0.67 | Improved |
| replace conventional camera with prototype sensor | conventional camera · Real face verification dataset | AUC: 0.89 -> 0.59 | Improved |
| use learned sensor outputs instead of Kinect depth for action recognition | I3D on Kinect · NTURGBD120 cross-setup | top-1 accuracy (%): 79.1% -> 70.5% | Hurt |
| move from simulation to real prototype sensor | learned sensor H · NYUv2 / real captures | delta1 accuracy (%): 84.69% -> 83.73% | No clear effect |
Closest papers by meaning
- Privacy-Preserving Optics for Enhancing Protection in Face De-Identification (CVPR 2024 · similarity 0.84)
- Validating Privacy-Preserving Face Recognition under a Minimum Assumption (CVPR 2024 · similarity 0.84)
- Asymmetric Invertible Threat: Learning Reversible Privacy Defense for Face Recognition (NeurIPS 2026 · similarity 0.83)
- DepthVanish: Optimizing Adversarial Interval Structures for Stereo-Depth-Invisible Patches (NeurIPS 2025 · similarity 0.83)
- Disguise without Disruption: Utility-Preserving Face De-identification (AAAI 2024 · similarity 0.83)
- Enhancing Facial Privacy Protection via Weakening Diffusion Purification (CVPR 2025 · similarity 0.83)
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.
