|M.Sc Student||Golan Izhak|
|Subject||Deep Anomaly Detection using Geometric Transformations|
|Department||Department of Computer Science||Supervisor||Professor Ran El-Yaniv|
We consider the problem of anomaly detection in images, and present a new detection technique. Given a sample of images, all known to belong to a “normal” class (e.g., dogs), we show how to train a deep neural model that can detect out-of-distribution images (i.e., non-dog objects).
Classic approaches for solving the anomaly detection problem do not perform well on high-dimensional data (e.g., images). In order to mitigate this issue, most of the recent works employ some reconstruction mechanism to reduce the dimensionality of the data, and activate anomaly detection techniques on the low-dimensional representation.
In contrast to these methods, the main idea behind our scheme is to train a multi-class classification model to discriminate between dozens of geometric transformations applied on all the given “normal” images. The auxiliary expertise learned by the model generates feature detectors that effectively identify, at test time, anomalous images based on the softmax activation (i.e., the output of the classifier, which represents the probability that an input image belongs to each class) statistics of the classification model when applied on transformed images.
We present extensive experiments using the proposed detector, which indicate that our technique consistently improves all known algorithms by a wide margin. We also show how our method can be easily adapted to employ prior knowledge when the “normal” class is comprised of several different classes, and the training data is labeled.
In addition, we conduct experiments that provide intuition for the effectiveness of our method.