Anomaly Detection Based on Zero-Shot Outlier Synthesis and Hierarchical Feature Distillation

Abstract

Anomaly detection suffers from unbalanced data since anomalies are quite rare. Synthetically generated anomalies are a solution to such ill or not fully defined data. However, synthesis requires an expressive representation to guarantee the quality of the generated data. In this paper, we propose a two-level hierarchical latent space representation that distills inliers' feature-descriptors (through autoencoders) into more robust representations based on a variational family of distributions (through a variational autoencoder) for zero-shot anomaly generation. From the learned latent distributions, we select those that lie on the outskirts of the training data as synthetic-outlier generators. And, we synthesize from them, i.e., generate negative samples without seen them before, to train binary classifiers. We found that the use of the proposed hierarchical structure for feature distillation and fusion creates robust and general representations that allow us to synthesize pseudo outlier samples. And in turn, train robust binary classifiers for true outlier detection (without the need for actual outliers during training). We demonstrate the performance of our proposal on several benchmarks for anomaly detection.

OE-CTST: Outlier-Embedded Cross Temporal Scale Transformer for Weakly-supervised Video Anomaly Detection

“…For such a case, it is intuitive to learn the spatiotemporal cues of videos pertaining to one-class (i.e. normal) distribution which is extensively adopted in unsupervised approaches [29,31,32], but never used in weakly-supervised video anomaly detection. This is due to the existence of a large intra-class variance in spatio-temporal cues of the normal distribution that makes one-class methods ineffective.…”

Detecting anomalies from liquid transfer videos in automated laboratory setting

“…An approach to address data scarcity is zero-shot, one-shot, and any-shot learning ( Ravi and Larochelle, 2016 ; Snell et al, 2017 ; Sung et al, 2018 ; Doshi and Yilmaz, 2020a ; Lu et al, 2020 ; Rivera et al, 2020 ). Generally, different types of augmentations are used in this type of setting.…”

Few-Shot Object Detection: Application to Medieval Musicological Studies

“…Object detection, as a subtask of computer vision, has been the focus of tremendous research interest over the past few years, from traditional computer vision algorithms such as Viola Jones, and the Histogram of Oriented Gradients Detector, which are still commonly used in mobile applications for their speed and accuracy, to new deep-learning-based models [18][19][20] such as Yolo [17], RCNN [16], SSD [21], and others. We can split the deep learning-based object detection models into two subgroups: one-stage detectors and two-stage detectors.…”