MULDE: Multiscale Log-Density Estimation via Denoising Score Matching for Video Anomaly Detection

Benchmark Model Rank Results
anomaly-detection-on-chuk-avenueMULDE-object-centric-micro#1AUC: 94.3%
video-anomaly-detection-on-shanghaitechMULDE-object-centric-micro#1AUC: 86.7%
video-anomaly-detection-on-shanghaitechMULDE-frame-centric-micro#4AUC: 81.3%
video-anomaly-detection-on-ucsd-ped2MULDE-object-centric-micro#1AUC: 99.7%
anomaly-detection-on-shanghaitechMULDE-object-centric-micro#2AUC: 86.7%
anomaly-detection-on-shanghaitechMULDE-frame-centric-micro#18AUC: 81.3%
anomaly-detection-on-ucsd-ped2MULDE-object-centric-micro#2AUC: 99.7%
video-anomaly-detection-on-shanghaitech-video-anomaly-detection-standard-frame-levelMULDE#2Frame AUC: 86.7
video-anomaly-detection-on-ubnormalMULDE-frame-centric-micro-one-class-classification#2AUC: 72.8%
anomaly-detection-on-ubnormalMULDE-frame-centric-micro-one-class-classification#3AUC: 72.8%
anomaly-detection-in-surveillance-videos-on-ucf-crimeMULDE-frame-centric-micro-one-class-classification#14ROC AUC: 78.5%