AI Paper Atlas

Research map / Video / Video segmentation and tracking

Learning Anomalies with Normality Prior for Unsupervised Video Anomaly Detection

Haoyue Shi, Le Wang, Sanping Zhou, Gang Hua, Wei Tang

ECCV 2024 · Video › Video segmentation and tracking · approach: Weakly supervised video anomaly detection · 5 citations (OpenAlex)

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 changedWhereMeasuredEffect
use normality prior (video start and end) as labeled normal snippets instead of random (ablation)normality propagation · UCF-Crimetest AUC (%): 60.33 -> 75.99Improved
use normality prior instead of data-driven prior (ablation)normality propagation · ShanghaiTechtest AUC (%): 83.59 -> 85.96Improved
use propagation with temporally-modulated feature similarity instead of direct comparison (ablation)pseudo-label generation · ShanghaiTechAUC (%): 72.33 -> 85.96Improved
combine temporal and feature similarity instead of either alone (ablation)normality propagation · UCF-CrimeAUC (%): 63.01 -> 75.99Improved
add temporal-only similarity propagation versus direct comparison on UCF-Crime (ablation)normality propagation · UCF-CrimeAUC (%): 71.22 -> 57.99Hurt
add loss re-weighting (ablation)normality propagation · UCF-CrimeAUC (%): 75.99 -> 76.64Improved
vary number of high-confidence normal videos h from 0.1 to 0.5 (ablation)loss re-weighting · ShanghaiTech and UCF-CrimeAUC (%): 86.46 -> 86.24 (ShanghaiTech)No clear effect
use I3D features instead of ResNext with this methodproposed method · UCF-CrimeAUC (%): 80.02 vs C2FPL 80.65Mixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy