Research map / 3D vision / Human pose estimation
Learning Privacy-Preserving Optics for Human Pose Estimation
Carlos Hinojosa, Juan Carlos Niebles, Henry Arguello
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 the first 20 VGG-19 backbone layers (versus no fine-tuning, 350 Zernike) (ablation) | PP-OPPS · COCO val2017 ablation | AP: no AP (did not converge) -> 0.302 | Improved |
| fine-tune 20 layers instead of 10 (ablation) | PP-OPPS · COCO val2017 ablation | AP: 0.263 -> 0.302 | Improved |
| fine-tune 40 layers instead of 20 (ablation) | PP-OPPS · COCO val2017 ablation | AP: 0.302 -> 0.251 | Hurt |
| increase Zernike coefficients from 15 to 350 (ablation) | PP-OPPS · COCO val2017 ablation | AP: 0.168 -> 0.302 | Improved |
| increase Zernike coefficients from 15 to 350 (ablation) | PP-OPPS · COCO val2017 ablation | PSNR: 16.692 -> 14.851 | Improved |
| replace defocus lens with the optimized multi-aberration lens | OpenPose with fixed optics · COCO val2017 | AP: 0.197 -> 0.302 | Improved |
| replace 32x32 low-resolution camera with the optimized lens | OpenPose with fixed optics · COCO val2017 | AP: 0.067 -> 0.302 | Improved |
| privacy-preserving degradation versus original OpenPose without privacy | OpenPose · COCO val2017 | AP: 0.421 -> 0.302 | Hurt |
| use privacy loss LP2 (face recognition similarity) instead of LP (ablation) | PP-OPPS · COCO val2017 | face recognition AUC / image distortion | Hurt |
| use privacy loss LP1 (scaled face loss) instead of LP (ablation) | PP-OPPS · COCO val2017 | HPE AP | Hurt |
Closest papers by meaning
- Learning a Dynamic Privacy-preserving Camera Robust to Inversion Attacks (ECCV 2024 · similarity 0.88)
- Privacy-Preserving Optics for Enhancing Protection in Face De-Identification (CVPR 2024 · similarity 0.87)
- Validating Privacy-Preserving Face Recognition under a Minimum Assumption (CVPR 2024 · similarity 0.85)
- LPSNet: End-to-End Human Pose and Shape Estimation with Lensless Imaging (CVPR 2024 · similarity 0.85)
- ViP: A Differentially Private Foundation Model for Computer Vision (ICML 2024 · similarity 0.85)
- Privacy-Shielded Image Compression: Defending Against Exploitation from Vision-Language Pretrained Models (ICML 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.
