Computer Vision News - August 2021

Hoel Kervadec 19 to fully realize all the possibilities that were opened up by it.” The paper focuses on 2D segmentation , but it can be easily generalized to 3D . One example would be computing the 3D descriptors for the minimum and maximum size of a heart in its cardiac cycle, and the diameter of the different ventricles at those two extremes, meaning that any point in the cardiac cycle will fall between those two bounds. You then have a valid description which you could reuse to supervise future scans for that patient without needing extra annotation . Originally from France , Hoel wanted to travel, and moved to Canada for his PhD when he was offered the opportunity to join Ismail Ben Ayed ’s group. He is moving to a new postdoc position in Marleen de Bruijne ’s group at Erasmus MC in September. Hoel says it was only after submitting this work that he fully appreciated its potential impact. “ What we show about generalization in 3D and defining bounds per patient would be valid for any scan in the future. It was a bit risky to submit it as we did not have a state-of-the-art result, but rather a different way to think about the problem and to make sense of the existing descriptors in the literature. It opens up so many possibilities for new research . ” Best of M I D L