Research map / Video / Video segmentation and tracking
Privacy-Aware Video Anomaly Detection through Orthogonal Subspace Projection
Lei Wang, Wenxiang Diao, Andrew Busch, Jun Zhou, Yongsheng Gao
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 G-OPL/OPL layers to MGFN | MGFN · ShanghaiTech (main) | frame-level AUC (%): 75.3 -> 83.7 | Improved |
| add G-OPL/OPL layers to RTFM | RTFM · UCF-Crime (main) | frame-level AUC (%): 74.3 -> 78.3 | Improved |
| add G-OPL to TEVAD | TEVAD · ShanghaiTech (main) | frame-level AUC (%): 98.1 -> 98.2 | No clear effect |
| add G-OPL instead of OPL alone | RTFM · ShanghaiTech (privacy-method comparison) | FPD (identity leakage): 0.30 -> 0.22 | Improved |
| add input masking (blur) to RTFM | RTFM · ShanghaiTech (privacy-method comparison) | frame-level AUC (%): 96.8 -> 93.5 | Hurt |
| add differential privacy noise to RTFM | RTFM · ShanghaiTech (privacy-method comparison) | frame-level AUC (%): 96.8 -> 95.8 | Hurt |
| add G-OPL to RTFM instead of GRL adversarial | RTFM · ShanghaiTech (privacy-method comparison) | frame-level AUC (%): 95.9 -> 97.3 | Improved |
| increase ranks k_OPL and k_G-OPL from 4 to 128 (ablation) | OPL + G-OPL · ShanghaiTech validation | AUC (%): 97.3 -> 87.9 | Hurt |
| use two G-OPL layers instead of one early layer (ablation) | G-OPL · ShanghaiTech (layer count) | FPD: 0.28 -> 0.22 | Improved |
| use three G-OPL layers instead of one early layer (ablation) | G-OPL · ShanghaiTech (layer count) | privacy-per-GFLOP: 0.233 -> 0.073 | Hurt |
| suppress face attribute with G-OPL (ablation) | G-OPL · ShanghaiTech (attribute-wise) | AUC (%): 83.8 -> 85.3 | Improved |
| suppress gait attribute with G-OPL (ablation) | G-OPL · ShanghaiTech (attribute-wise) | AUC (%): 83.8 -> 83.2 | 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.
- MGFN — MGFN: Magnitude-Contrastive Glance-and-Focus Network for Weakly-Supervised Video Anomaly Detection (AAAI 2023) · better on 2 of 2 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
- VPD-100K: Towards Generalizable and Fine-grained Visual Privacy Protection (ICML 2026 · similarity 0.86)
- TeD-SPAD: Temporal Distinctiveness for Self-Supervised Privacy-Preservation for Video Anomaly Detection (ICCV 2023 · similarity 0.86)
- Collaborative Learning of Anomalies with Privacy (CLAP) for Unsupervised Video Anomaly Detection: A New Baseline (CVPR 2024 · similarity 0.84)
- Disguise without Disruption: Utility-Preserving Face De-identification (AAAI 2024 · similarity 0.82)
- Validating Privacy-Preserving Face Recognition under a Minimum Assumption (CVPR 2024 · similarity 0.82)
- FrameShield: Adversarially Robust Video Anomaly Detection (NeurIPS 2025 · 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.
