Research map / Video / Action recognition
Less Static, More Private: Towards Transferable Privacy-Preserving Action Recognition by Generative Decoupled Learning
Zhi-Wei Xia, Kun-Yu Lin, Yuan-Ming Li, Wei-Jin Huang, Xian-Tuo Tan, Wei-Shi Zheng
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 GenPriv decoupled learning versus best prior method | privacy-preserving action recognition · TP-HMDB->TP-UCF | Top-1 accuracy (%): 72.22 -> 87.91 | Improved |
| add GenPriv versus best prior method | privacy-preserving action recognition · TP-UCF->TP-HMDB | Top-1 accuracy (%): 74.26 -> 80.55 | Improved |
| add DANN domain adaptation on top of privacy-preserved data | VITA / SPAct · TP-HMDB->TP-UCF | Top-1 accuracy (%): 65.50 -> 67.37 (VITA), 72.22 -> 63.39 (SPAct) | Mixed |
| add temporal alignment loss with decoupled features (prior alignment) (ablation) | GenPriv · TP-UCF/TP-HMDB transfer (Table 3 first column) | Top-1 accuracy (%): 80.56 -> 87.91 | Improved |
| align after decoupling rather than before (ablation) | GenPriv · TP-UCF/TP-HMDB transfer (Table 3 first column) | Top-1 accuracy (%): 87.91 -> 48.46 | Hurt |
| add mutual information loss (ablation) | GenPriv · TP-UCF->TP-HMDB | Top-1 accuracy (%): 71.94 -> 80.55 | Improved |
| add spatial consistency loss to decoupled latent space (ablation) | GenPriv · TP-UCF (privacy) | privacy cMAP (%): 70.14 -> 67.42 | Improved |
| add spatial consistency loss to decoupled latent space (ablation) | GenPriv · TP-HMDB (privacy) | privacy cMAP (%): 63.09 -> 64.84 | Hurt |
| use hybrid static+dynamic reconstruction instead of dynamic only (ablation) | GenPriv · TP-HMDB (same-dataset) | Top-1 accuracy (%): 79.17 -> 88.05 | Improved |
| GenPriv privacy leakage versus other methods | GenPriv · TP-HMDB->TP-UCF | privacy F1: 0.519 vs 0.517 (VITA+DANN) | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SPAct — SPAct: Self-Supervised Privacy Preservation for Action Recognition (CVPR 2022) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- STPrivacy: Spatio-Temporal Privacy-Preserving Action Recognition (ICCV 2023 · similarity 0.91)
- Joint Attribute and Model Generalization Learning for Privacy-Preserving Action Recognition (NeurIPS 2023 · similarity 0.90)
- Privacy-Preserving Deep Action Recognition: An Adversarial Learning Framework and A New Dataset (TPAMI 2020 · similarity 0.90)
- Selective Interpretable and Motion Consistent Privacy Attribute Obfuscation for Action Recognition (CVPR 2024 · similarity 0.88)
- ReGenHuman: Re-Generating Human Appearances for Realistic Full-Body Video Anonymization (NeurIPS 2026 · similarity 0.84)
- MultiPriv: Benchmarking Individual-Level Privacy Reasoning in Vision-Language Models (ICML 2026 · 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.
