Using a technique called semantic manipulation, the scene starts by building a label map image and then sending that through an A.I. network to swap the labeled scene with imaging data it has been trained in to recreate the realistic image. This greatly helps reduce the effort and cost of 3D modeling in games and capturing training data for self-driving cars. Machines teaching machines, well, almost.
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.