Computer Vision News - August 2022

34 Best Paper MIDL 1st runner-up a nice example of where a good idea translated into practice and worked from the start – with a few tweaks here and there, of course. ” Thinking about what comes next, Tamaz tells us he would like to consider the different applications of this generative model. It has been shown to work well on reconstruction tasks, but could this shape analysis be applied to other scenarios in the medical domain, or even outside of it? “ The kind of data we use for training occurs naturally in the medical domain, ” he responds. “ In computer vision, you typically would have sparse measurements from LiDAR scanners , for example, where you get point clouds from only one side of the object. Certainly, it has been applied there, but it’s a different setting. ” Before CNNs and deep learning, people built shape prior models to solve the segmentation task because image- based models were very simple. Specific statistical shape models were used extensively to regularize the segmentation process because it was hard to make sense of those images. It used to be tedious to build statistical shape models, with manual