Research map / Video / Video segmentation and tracking
Self-Supervised Sparse Representation for Video Anomaly Detection
Jhih-Ciang Wu, He-Yen Hsieh, Ding-Jie Chen, Chiou-Shann Fuh, Tyng-Luh Liu
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 de-Normal module with averaged normal feature (ablation) | S3R baseline · UCF-Crime (weakly-supervised) | AUC: 83.42 -> 84.19 | Improved |
| add de-Normal module with averaged normal feature (ablation) | S3R baseline · ShanghaiTech (weakly-supervised) | AUC: 96.97 -> 97.28 | Improved |
| use learned task-specific dictionary DT instead of averaged feature (ablation) | S3R · UCF-Crime (weakly-supervised) | AUC: 84.19 -> 85.99 | Improved |
| use task-specific DT instead of universal DU for reconstruction (ablation) | S3R · UCF-Crime (weakly-supervised) | AUC: 85.00 -> 85.99 | Improved |
| use task-specific DT instead of universal DU (ablation) | S3R · ShanghaiTech (weakly-supervised) | AUC: 97.47 -> 97.48 | No clear effect |
| use universal dictionary DU for pseudo anomaly generation | S3R · UCF-Crime (one-class) | AUC: 77.15 -> 79.58 | Improved |
| use universal dictionary DU for pseudo anomaly generation | S3R · ShanghaiTech (one-class) | AUC: 79.89 -> 80.47 | Improved |
| replace 25% anomaly and 25% normal snippets vs 25% anomaly only (ablation) | S3R · ShanghaiTech (one-class), pseudo snippet ratio | AUC: 79.59 -> 80.47 | Improved |
| use 12.5% anomaly ratio instead of 25% (ablation) | S3R · ShanghaiTech (one-class), pseudo snippet ratio | AUC: 80.47 -> 60.43 | Hurt |
| set en-Normal channel reduction to 25% instead of 50% (ablation) | S3R · ShanghaiTech (one-class), en-Normal channel reduction | AUC: 70.38 -> 80.47 | Improved |
| set de-Normal MLP reduction to 6.25% instead of 25% (ablation) | S3R · ShanghaiTech (one-class), de-Normal channel reduction | AUC: 66.14 -> 80.47 | Improved |
| S3R vs best prior one-class method on UCF-Crime | S3R · UCF-Crime (one-class) | AUC: 72.70 -> 77.15 | Improved |
Closest papers by meaning
- Generalizing Single-Frame Supervision to Event-Level Understanding for Video Anomaly Detection (NeurIPS 2025 · similarity 0.90)
- Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · similarity 0.89)
- VADTree: Explainable Training-Free Video Anomaly Detection via Hierarchical Granularity-Aware Tree (NeurIPS 2025 · similarity 0.87)
- VadCLIP: Adapting Vision-Language Models for Weakly Supervised Video Anomaly Detection (AAAI 2024 · similarity 0.87)
- Open-Vocabulary Video Anomaly Detection (CVPR 2024 · similarity 0.87)
- Text Prompt with Normality Guidance for Weakly Supervised Video Anomaly Detection (CVPR 2024 · similarity 0.87)
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.
