Pretty impressive. I wonder though why it was necessary to put the point map of the second image into the coordinate frame of the first. Isn’t it all the same from the point of the neural net?