Research map / Video / Action recognition
SPAct: Self-Supervised Privacy Preservation for Action Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 supervised privacy labels with self-supervised contrastive privacy removal | SPAct · UCF101 | Top-1 accuracy: 62.10 -> 62.03 | No clear effect |
| Replace supervised privacy labels with self-supervised contrastive privacy removal | SPAct · VISPR1 | cMAP: 55.32 -> 57.43 | Hurt |
| Use self-supervised rather than supervised anonymization | SPAct · VISPR1→VISPR2 | cMAP: 49.6 -> 47.1 | Helped |
| Use self-supervised rather than supervised anonymization | SPAct · P-HVU Scenes→Objects | cMAP: 6.43 -> 1.42 | Helped |
| Remove the privacy-removal branch (ablation) | SPAct · VISPR1 | cMAP: 57.4 -> 64.1 | Hurt |
| Freeze the spatial SSL privacy model instead of updating it during minimax optimization (ablation) | SPAct · VISPR1 | cMAP: 57.4 -> 62.2 | Hurt |
| Use spatio-temporal instead of spatial SSL for privacy removal (ablation) | SPAct · UCF101 | Top-1 accuracy: 62.1 -> 56.4 | Hurt |
| Sample positive frames more than 64 frames apart (ablation) | SPAct · VISPR1 | cMAP: 57.4 -> 58.7 | Hurt |
| Sample positive frames less than 8 frames apart (ablation) | SPAct · VISPR1 | cMAP: 57.4 -> 57.1 | No clear effect |
| Replace NT-Xent with RotNet (ablation) | SPAct · UCF101 | Top-1 accuracy: 62.1 -> 58.1 | Hurt |
| Replace NT-Xent with MoCo (ablation) | SPAct · VISPR1 | cMAP: 57.4 -> 57.1 | No clear effect |
| Use a stronger auxiliary privacy model, combining ResNet-50 and MobileNetV1 (ablation) | SPAct · VISPR1 | cMAP: 57.43 -> 56.20 | Helped |
Closest papers by meaning
- Privacy Beyond Pixels: Latent Anonymization for Privacy-Preserving Video Understanding (ICLR 2026 · similarity 0.89)
- Joint Attribute and Model Generalization Learning for Privacy-Preserving Action Recognition (NeurIPS 2023 · similarity 0.88)
- Privacy-Preserving Video Classification with Convolutional Neural Networks (ICML 2021 · similarity 0.87)
- Selective Interpretable and Motion Consistent Privacy Attribute Obfuscation for Action Recognition (CVPR 2024 · similarity 0.86)
- StegaVAR: Privacy-Preserving Video Action Recognition via Steganographic Domain Analysis (AAAI 2026 · similarity 0.85)
- MaSS: Multi-attribute Selective Suppression for Utility-preserving Data Transformation from an Information-theoretic Perspective (ICML 2024 · similarity 0.82)
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.