Computer Vision News - July 2021

Presentation 36 Best of CVPR 2021 those messages in the convolutional scheme. “ With the B-rep structure, we actually have a way to do better than that, ” Joe explains. “ We can describe any neighboring edge, face or coedge using this idea of what we topological walks. A topological walk is a list of instructions which tell us how to move from one coedge to another. If we think of the instruction list to go to the next coedge and then to its mating coedge and then to jump to the parent face, then using this sequence of instructions, we have a way to address this face from this coedge in a completely unique way. This sort of thing is just not possible with a graph network, where if I want to talk about this node in the graph, all I can do is I can talk about the neighboring nodes. I can’t choose a specific one. ” The BRepNet architecture is designed specifically to take advantage of the additional capability stored in the B - rep models. Using this set of topological walks, you can describe neighboring faces and all neighboring edges and then take the data of whatever feature vectors you choose