Adversarially Learned One-Class Classifier for Novelty Detection

Abstract

Novelty detection is the process of identifying the observation(s) that differ in some respect from the training observations (the target class). In reality, the novelty class is often absent during training, poorly sampled or not well defined. Therefore, one-class classifiers can efficiently model such problems. However, due to the unavailability of data from the novelty class, training an end-to-end deep network is a cumbersome task. In this paper, inspired by the success of generative adversarial networks for training deep models in unsupervised and semi-supervised settings, we propose an end-to-end architecture for one-class classification. Our architecture is composed of two deep networks, each of which trained by competing with each other while collaborating to understand the underlying concept in the target class, and then classify the testing samples. One network works as the novelty detector, while the other supports it by enhancing the inlier samples and distorting the outliers. The intuition is that the separability of the enhanced inliers and distorted outliers is much better than deciding on the original samples. The proposed framework applies to different related applications of anomaly and outlier detection in images and videos. The results on MNIST and Caltech-256 image datasets, along with the challenging UCSD Ped2 dataset for video anomaly detection illustrate that our proposed method learns the target class effectively and is superior to the baseline and state-of-the-art methods.

Self-Supervised Representation Learning via Neighborhood-Relational Encoding

“…We evaluate our method on the UCSD Ped2 dataset [8], which is a popular dataset for this task. We follow the evaluation criteria of [51]. Similar to [51], the frame-level accuracy is reported as the performance metric.…”

“…We follow the evaluation criteria of [51]. Similar to [51], the frame-level accuracy is reported as the performance metric. In frame-level measure, a frame is considered as anomaly, if at least one of its pixels is detected as anomaly.…”

Anomaly Detection in Video Sequence With Appearance-Motion Correspondence

“…This bad estimation significantly affected the error map though the three cars running on other way were correctly determined. The results may thus be im- Method Belleview Train GANomaly [2] 0.735 0.194 AEs + local feature [35] 0.748 0.171 AEs + global feature [35] 0.776 0.216 ALOCC D(X) [40] 0.734 0.182 ALOCC D(R(X)) [40] 0.805 0.237 Our proposed method 0.751 0.490 SSIM on appearance stream 0.830 0.798 Table 3. The average precision of frame-level anomaly detection on the Traffic-Belleview and Traffic-Train datasets.…”

Fast Distance-Based Anomaly Detection in Images Using an Inception-Like Autoencoder

“…Deep approaches to anomaly detection for image data often use a convolutional autoencoder (CAE) which include convolutional layers in the AE architecture [24,31]. Another line of work uses Generative Adversarial Networks (GAN) for this task [8,27,29]. This two-step process is also used to make the density estimation task easier by learning low-dimensional representations.…”