CVPR Daily - Tuesday

DAILY T u e s d a y Boyang Deng 7 The idea for this work grew out of a discussion between Boyang and Andrea Tagliasacchi , a senior research scientist at Google Brain who Boyang mentioned “is a hands-on mentor teachingme how to do good research”. They had some issues using meshes to represent shapes on previous projects and Andrea proposed that convexity may help. By using implicit representation for the meshes, it might be easier for machine learning models to learn to represent shapes, instead of using human effort to design representations or topologies. They created a hybrid representation. “To represent convex shapes, we use the idea that any convex can be represented by the intersection of a lot of half-spaces or half-planes in 3D.” “Hybrid is the most exciting word to me!” he exclaims. “During training, this representation is implicit. We can treat it just as a function. That means it’s not that different from traditional machine learning. We just cast the shape representation problem to classification and apply the machine