Research map / Video / Action recognition
Privacy-Preserving Deep Action Recognition: An Adversarial Learning Framework and A New Dataset
Zhenyu Wu, Haotao Wang, Zhaowen Wang, Hailin Jin, Zhangyang Wang
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 |
|---|---|---|---|
| replace GRL with negative-entropy objective (Ours-Entropy M=1) | adversarial anonymization · SBU | privacy budget accuracy A_B (%): 85.4 -> 73.6 | Improved |
| add model restarting to Ours-Entropy (M=1) (ablation) | Ours-Entropy · SBU | privacy budget accuracy A_B (%): 73.6 -> 60.5 | Improved |
| increase ensemble size from M=1 to M=8 with restarting (ablation) | Ours-Entropy · SBU | privacy budget accuracy A_B (%): 60.5 -> 47.7 | Improved |
| add model restarting to K-Beam (K=2) (ablation) | Ours-K-Beam · SBU | privacy budget accuracy A_B (%): 87.7 -> 80.6 | Improved |
| add model restarting to GRL (ablation) | GRL · SBU | privacy budget accuracy A_B (%): 85.4 -> 84.3 | Improved |
| add model restarting to Ours-Entropy (M=4) on SBU action accuracy (ablation) | Ours-Entropy · SBU | action accuracy A_T (%): 82.7 -> 78.0 | Hurt |
| increase K-Beam beams from K=4 to K=8 (ablation) | Ours-K-Beam · SBU | privacy budget accuracy A_B (%): 88.4 -> 80.2 | Mixed |
| use naive downsampling rate 16 instead of 1 | Downsample · SBU | action accuracy A_T (%): 88.8 -> 64.4 | Hurt |
| apply Ours-Entropy with ensemble and restarting | Ours-Entropy · UCF101/VISPR | privacy budget accuracy A_B (%): 45.5 -> 34.5 | Improved |
| apply Ours-Entropy M=4 without retraining | Ours-Entropy · PA-HMDB51 | privacy budget accuracy A_B (%): 77.3 -> 71.6 | Improved |
| blur or blacken the body instead of the face | empirical obfuscation · SBU | privacy budget accuracy A_B (%): 98.9 -> 21.4 | Mixed |
| apply learned anonymization, human raters | f_A · PA-HMDB51 human study | macro-F1 of privacy attribute prediction: 0.97 -> 0.38 | Improved |
Closest papers by meaning
- Towards Privacy-Preserving Visual Recognition via Adversarial Training: A Pilot Study (ECCV 2018 · similarity 0.95)
- Learning to Anonymize Faces for Privacy Preserving Action Detection (ECCV 2018 · similarity 0.92)
- Privacy Beyond Pixels: Latent Anonymization for Privacy-Preserving Video Understanding (ICLR 2026 · similarity 0.92)
- STPrivacy: Spatio-Temporal Privacy-Preserving Action Recognition (ICCV 2023 · similarity 0.91)
- Selective Interpretable and Motion Consistent Privacy Attribute Obfuscation for Action Recognition (CVPR 2024 · similarity 0.88)
- Balancing Privacy and Performance: A Many-in-One Approach for Image Anonymization (AAAI 2025 · similarity 0.86)
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.
