Computer Vision News - November 2021

50 AI Research Paper Hi everyone, how are you doing? Let’s have another article to read. This month’s review is “Anomaly detection in medical imaging with deep perceptual autoencoders” by Nina Shvetsova et al. This is a pre-print which was uploaded this month, but it already has 15 citations and you can easily find it on Arxiv, to read it in full. Anomaly detection in medical imaging with deep perceptual autoencoders Anomaly Detection Anomaly detection is a task with significance, especially in the deployment of machine learning models. The knowledge of a “a normal” data sample would be used to compare -in a sense of a ground truth- to an “abnormal” one. To identify less often occurrences is another application where anomaly detection is useful and with the method proposed in this paper, the efficacy of autoencoders for anomaly detection is utilized. The authors compared here the three strongest SOTA anomaly detection methods in two challenging medical tasks: Chest X-rays and H&E-stained histological images . In the manner of open science (Richard Stallman would call the free, oh well!), the source code of all our experiments to facilitate the development of anomaly detection in medical imaging is shared online. The solution was sequentially compared with both datasets with SOTA methods which were outperformed. Even though Computer Vision News and Medical Imaging News focus on computer vision and medical learning, there is a wide range of fields where anomaly detection