Computer Vision News - August 2021

MIDL Presentation 24 Self-Supervised Out-of- Distribution Detection for Cardiac CMR Segmentation Camila González is a second-year PhD candidate in the Medical and Environmental Computing Lab at the Technical University of Darmstadt in Germany, under the supervision of Anirban Mukhopadhyay. Her paper explores out-of-distribution detection, and she spoke to us ahead of her short oral discussion session. Camila begins by telling us the wider research topic for her PhD is continual learning for multi-domain data . Specifically for data that comes from different institutions, or data that was acquired with different pieces of equipment, such as CT or MRT data acquired by machines from different vendors. When models are trained on this multi-domain data, they often do not generalize well to data from other places or vendors. This causes deep learning models to fail silently . For quality assurance, it is important to identify these out-of-distribution samples for which the trained model is unsuitable. “ Regardless of whether you have a model that aims to generalize well to other domains, you still need methods to detect when a sample is out-of- distribution,” Camila explains. “That’s why in this paper we look at self-supervised models and how to use self-supervision loss calculated during testing to detect out-of-distribution samples . There are many different Camila Gonzalez Best of M I D L