DeepMind, Google’s artificial intelligence subsidiary in London, has developed a self-training vision computer that generates “a full 3D model of a scene from just a handful of 2D snapshots ...