Computer Vision News - June 2023

24 Oral Paper at MIDL 2023 supervised models to medical images, focusing specifically on chest X-rays. She aims to train thesemodels without explicit labels, using self-supervised learning techniques such as pseudo-learning and contrastive frameworks. The key element that drives the success of self-supervised learning is the use of augmentations . Augmentations involve applying various computer vision transformations to an image, such as Gaussian blurring or random resized cropping . Intheself-supervisedframework, the image is transformed into two different augmentations, and their similarity or dissimilarity is assessed, which is the NanditaBhaskharhas just completed her PhD in the Department of Electrical Engineering at Stanford University. With the success of large language models like ChatGPT, there is an avid interest in doing the same thing but for medical images . However, medical images are a completely different beast. They have domain-specific properties that differentiate them from natural images, such as those of cats and dogs. In this upcoming MIDL paper, co-written with Rogier van der Sluijs , Nandita addresses the need to apply large self- EXPLORING IMAGE AUGMENTATIONS FOR SIAMESE REPRESENTATION LEARNING WITH CHEST X-RAYS