GLAD: A Global-to-Local Anomaly Detector

Artola, Aitor; Kolodziej, Yannis; Morel, Jean‐Michel; Ehret, Thibaud

doi:10.1109/wacv56688.2023.00546

Cited by 4 publications

(3 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Implementation details. We trained a GMM on each sequence setting K = 1000 initial Gaussian distributions and we let GLAD remove unnecessary Gaussians, as proposed by its authors [34]. For training, we selected the first few hundred subsequent frames of each scene without any obvious anomalies.…”

Section: Methodsmentioning

confidence: 99%

“…We model the extracted deep representations with a mixture of Gaussians to assess the likelihood of an image patch to be a part of the background. For that we extend the GLAD [34] framework to the background modeling of videos. This requires no dense annotations, only a selection of training frames with none or few anomalies present, thus we label our approach as weakly supervised.…”

Section: Background Feature Modelingmentioning

confidence: 99%

“…PaDiM, proposed by Defard et al [33], is a framework that models patches of a DNN feature map with a Gaussian model for anomaly detection and localization. Artola et al [34] generalized this attempt with GLAD, a method that learns a robust GMM globally, and then localizes the learned Gaussians with a spatial weight map. Modeling spatial features in a high-dimensional space has shown promising results at recognizing complex patterns.…”

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Statistical modeling of deep features reduces false alarms in video change detection

Bou,

Artola,

Ehret

et al. 2024

Preprint

Self Cite

View full text Add to dashboard Cite

Detecting relevant changes is a fundamental problem of video surveillance. Because of the high variability of data and the difficulty of properly annotating changes, unsupervised methods dominate the field. Arguably one of the most critical issues to make them practical is to reduce their false alarm rate. In this work, we develop a method-agnostic weakly supervised a-contrario validation process, based on high dimensional statistical modeling of deep features, to reduce the number of false alarms of any change detection algorithm. We also raise the insufficiency of the conventionally used pixel-wise evaluation, as it fails to precisely capture the performance needs of most real applications. For this reason, we complement pixel-wise metrics with object-wise metrics and evaluate the impact of our approach at both pixel and object levels, on six methods and several sequences from different datasets. Experimental results reveal that the proposed a-contrario validation is able to largely reduce the number of false alarms at both pixel and object levels.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Background Feature Modelingmentioning

confidence: 99%