Research map / Video / Video segmentation and tracking
Unmasking the Abnormal Events in Video
Radu Tudor Ionescu, Sorina Smeureanu, Bogdan Alexe, Marius Popescu
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 unmasking (ablation) | VGG-f conv5 features · Avenue (5 test videos, preliminary) | frame-level AUC (%): 78.6% -> 81.1% | Improved |
| divide frames into 2x2 spatial bins (ablation) | VGG-f conv5 with unmasking · Avenue (5 test videos, preliminary) | frame-level AUC (%): 81.1% -> 82.5% | Improved |
| use conv5 instead of fc7 features (ablation) | VGG-f without unmasking · Avenue (5 test videos, preliminary) | frame-level AUC (%): 78.3% -> 78.6% | Improved |
| increase stride from 1 to 10 (ablation) | VGG-f conv5, 2x2 bins, unmasking · Avenue (5 test videos, preliminary) | frame-level AUC (%): 82.5% -> 82.0% | No clear effect |
| increase stride from 1 to 10 (speed) (ablation) | VGG-f conv5, 2x2 bins, unmasking · Avenue (5 test videos, preliminary) | prediction speed (FPS): 9.4 -> 78.1 | Improved |
| add late fusion of 3D gradients with conv5 (ablation) | unmasking framework · Avenue (5 test videos, preliminary) | frame-level AUC (%): 79.8% -> 82.6% | Improved |
| add late fusion of motion and appearance | unmasking with 3D gradients · Subway Entrance gate | frame-level AUC (%): 71.3% -> 70.6% | Hurt |
| use unmasking framework instead of Del Giorno et al. | unsupervised anomaly detection · Avenue | frame-level AUC (%): 78.3% -> 80.6% | Improved |
| use unmasking framework instead of Del Giorno et al. | unsupervised anomaly detection · UMN (all scenes) | frame-level AUC (%): 91.0% -> 95.1% | Improved |
| use CNN conv5 appearance features instead of 3D gradient motion features | unmasking framework · UMN (scene 2) | frame-level AUC (%): 84.9% -> 86.5% | Improved |
| use 3D gradient motion features instead of conv5 appearance features | unmasking framework · Subway Exit gate | frame-level AUC (%): 84.7% -> 86.3% | Improved |
Closest papers by meaning
- Object-Centric Auto-Encoders and Dummy Anomalies for Abnormal Event Detection in Video (CVPR 2019 · similarity 0.89)
- A Background-Agnostic Framework with Adversarial Training for Abnormal Event Detection in Video (TPAMI 2021 · similarity 0.88)
- Learning Anomalies with Normality Prior for Unsupervised Video Anomaly Detection (ECCV 2024 · similarity 0.86)
- Learning Event Completeness for Weakly Supervised Video Anomaly Detection (ICML 2025 · similarity 0.86)
- Learning to Tell Apart: Weakly Supervised Video Anomaly Detection via Disentangled Semantic Alignment (AAAI 2026 · similarity 0.86)
- Self-Distilled Masked Auto-Encoders are Efficient Video Anomaly Detectors (CVPR 2024 · similarity 0.85)
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.
