Computer Vision News - November 2022

14 Best Paper Honorable Mention Shape optimization is everywhere . It is a critical component in manufacturing, engineering, and many other fields and is used to model aircraft, cars, and satellites. One approach to optimizing these shapes is by using explicit surface representations, which means performing shape optimization with triangle meshes and point clouds, for example. “ The downside of using explicit representations in geometry optimization is the problem of controlling the topology of the surfaces. Topology is what the connectivity of the shape looks like , ” Ishit tells us. “ The topology of a sphere and a donut are different because a donut has a hole, whereas a sphere doesn’t. That’s the classic example. If you use explicit surface representations, going from a sphere to a donut is difficult because it’s a discontinuous operation. You need to create a hole. That’s hard to do with standard optimization. ” Recovering or optimizing geometry is a very common process. It is called an inverse problem, where you know the Ishit Mehta is a PhD student at the University of California San Diego under the supervision of Manmohan Chandraker and Ravi Ramamoorthi. His paper proposing a theoretical framework to analyze explicit and implicit surface representations in shape optimization has just won a Best Paper Honorable Mention award at ECCV 2022. He speaks to us about his work. A LEVEL SET THEORY FOR NEURAL IMPLICIT EVOLUTION UNDER EXPLICIT FLOWS HONORABLE M E N T I O N AWARD ECCV