Research map / Video / Video segmentation and tracking
Weakly-Supervised Video Anomaly Detection With Robust Temporal Feature Magnitude Learning
Yu Tian, Guansong Pang, Yuanhong Chen, Rajvinder Singh, Johan W. Verjans, Gustavo Carneiro
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 feature magnitude learning (FM) to the MIL baseline (ablation) | MIL baseline · ShanghaiTech | AUC: 85.96 -> 92.99 | Improved |
| add feature magnitude learning (FM) to the MIL baseline (ablation) | MIL baseline · UCF-Crime | AUC: 77.39 -> 81.28 | Improved |
| add pyramid of dilated convolutions (PDC) (ablation) | MIL baseline · ShanghaiTech | AUC: 85.96 -> 89.21 | Improved |
| add temporal self-attention (TSA) (ablation) | MIL baseline · ShanghaiTech | AUC: 85.96 -> 91.73 | Improved |
| add PDC (ablation) | MIL baseline · UCF-Crime | AUC: 77.39 -> 79.32 | Improved |
| add TSA (ablation) | MIL baseline · UCF-Crime | AUC: 77.39 -> 78.96 | Improved |
| add TSA alongside PDC (ablation) | MIL baseline with PDC · ShanghaiTech | AUC: 89.21 -> 92.32 | Improved |
| add PDC and TSA to the FM model (full RTFM) (ablation) | MIL baseline with FM · ShanghaiTech | AUC: 92.99 -> 97.21 | Improved |
| add PDC and TSA to the FM model (full RTFM) (ablation) | MIL baseline with FM · UCF-Crime | AUC: 81.28 -> 84.30 | Improved |
| add FM to the PDC+TSA model (ablation) | MTN with MIL · UCF-Crime | AUC: 82.12 -> 84.30 | Improved |
| replace Sultani MIL with RTFM (I3D features) (ablation) | Sultani et al. · UCF-Crime | AUC: 77.92 -> 84.30 | Improved |
| reduce abnormal training videos from 63 to 25 and use RTFM instead of MIL (ablation) | RTFM vs Sultani · ShanghaiTech | AUC: about 3% higher using 25 vs 63 videos | Improved |
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.
- AR-Net — AR-Net: Adaptive Frame Resolution for Efficient Action Recognition (ECCV 2020) · 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
- TLMA: Mitigating the Impact of Weakly Labeled Information for Video Anomaly Detection (CVPR 2026 · similarity 0.91)
- Prompt-Enhanced Multiple Instance Learning for Weakly Supervised Video Anomaly Detection (CVPR 2024 · similarity 0.89)
- Learning to Tell Apart: Weakly Supervised Video Anomaly Detection via Disentangled Semantic Alignment (AAAI 2026 · similarity 0.89)
- MIST: Multiple Instance Self-Training Framework for Video Anomaly Detection (CVPR 2021 · similarity 0.89)
- Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · similarity 0.88)
- VadCLIP: Adapting Vision-Language Models for Weakly Supervised Video Anomaly Detection (AAAI 2024 · similarity 0.88)
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.
