Research map / Video / Action recognition
STPrivacy: Spatio-Temporal Privacy-Preserving Action Recognition
Ming Li, Xiangyu Xu, Hehe Fan, Pan Zhou, Jun Liu, Jia-Wei Liu, Jiahe Li, Jussi Keppo, Mike Zheng Shou, Shuicheng Yan
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 |
|---|---|---|---|
| use video-level ViT tubelet sparsification plus anonymization instead of frame-level CNN transformation | privacy-preserving action recognition · VP-HMDB51 | Top-1 accuracy: 48.56 -> 50.73 | Improved |
| use STPrivacy instead of VITA | privacy-preserving action recognition · VP-UCF101 | Top-1 accuracy: 78.49 -> 82.55 | Improved |
| use STPrivacy on unseen actions | privacy-preserving action recognition · VP-HMDB51 to VP-UCF101 | Top-1 accuracy: 78.13 -> 81.04 | Improved |
| apply sparsification only (ablate anonymization block) (ablation) | STPrivacy · VP-HMDB51, VP-UCF101 | action-privacy trade-off | Improved |
| apply anonymization only (alpha=1.0) (ablation) | STPrivacy · VP-HMDB51, VP-UCF101 | action-privacy trade-off | Improved |
| increase tubelet keeping proportion alpha (ablation) | STPrivacy · VP-HMDB51, VP-UCF101 | action vs privacy recognition | Mixed |
| tubelet temporal length 1 instead of 2 (ablation) | STPrivacy · VP-HMDB51, VP-UCF101 | action recognition | Hurt |
| tubelet temporal length 4 instead of 2 (ablation) | STPrivacy · VP-HMDB51, VP-UCF101 | action recognition | Hurt |
| vary lambda_Action and lambda_Privacy weights (ablation) | STPrivacy · VP-HMDB51, VP-UCF101 | action-privacy trade-off | No clear effect |
| spatial downsampling for privacy removal | raw video · VP-HMDB51 | Top-1 accuracy: 51.44 -> 31.32 (4x) | Hurt |
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 7 of 7 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
- Less Static, More Private: Towards Transferable Privacy-Preserving Action Recognition by Generative Decoupled Learning (ICCV 2025 · similarity 0.91)
- Privacy-Preserving Deep Action Recognition: An Adversarial Learning Framework and A New Dataset (TPAMI 2020 · similarity 0.91)
- Joint Attribute and Model Generalization Learning for Privacy-Preserving Action Recognition (NeurIPS 2023 · similarity 0.90)
- Towards Privacy-Preserving Visual Recognition via Adversarial Training: A Pilot Study (ECCV 2018 · similarity 0.90)
- Selective Interpretable and Motion Consistent Privacy Attribute Obfuscation for Action Recognition (CVPR 2024 · similarity 0.89)
- EgoPrivacy: What Your First-Person Camera Says About You? (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.
