Computer Vision News - April 2019

15 SkelNetOn - CVPR 2019 organized through a research network called Women in Shape Modelling , sponsored by the Association for Women in Mathematics . Every few years it brings together women who are interested in shape modelling to work on open research problems together. At the workshop, there was one group working on medial axis and skeletons, and another working on deep learning, which gave the team the idea to join the two together. Kathryn , who is a professor at Occidental College, tells us that she has really enjoyed working on the challenge: “ The challenge has been challenging for us as well, but it’s been a lot of fun getting a peek behind the scenes of how these things get put together and all of the different moving parts. We have people in France, Germany, and across the United States, so just working with this really diverse team of creative and good-humored people has been great. ” Camilla , M.S. from Bergische Universität Wuppertal, adds: “I’m in the very early stage of my career, and probably family as well, so it’s been a great opportunity to meet so many great women in so many different steps of their careers, and work with them together on this great project. ” Géraldine , a professor from the University of Toulouse, echoed this sentiment: “ It was so much fun to start this working group and this challenge, and I think we’ve been ambitious. We are very happy to get some new ideas on this topic. ” What do they hope to achieve? Ilke , a research scientist at DeepScale, explains that geometric deep learning is an area that is not yet explored to its full potential: “ I think the ultimate outcome is some groundbreaking deep learning approach that has its own representation for shapes, and has its own convolutions and pooling and everything to understand shapes. What we expect from this workshop is to find that breakthrough deep learning architecture that can be used for shape understanding . ” Géraldine agrees: “ I think it’s a real challenge. There are all these devices to capture the world around us, like LiDAR scanners, reconstruction, set of points, and we know that deep learning is doing great for images. If this challenge can output a network that is able to learn a lot of things on this 3D data that we get through all of these devices, that would be really great. ” Challenge Computer Vision News “ If this challenge can output a network that is able to learn a lot of things on this 3D data that we get through all of these devices, that would be really great! ”