
float32 on 2048x2048x3 Tensor A generator knew how to dream of human faces. Nothing of frogs. A Resnet saw the dream and pushed back. It was adversarially trained. MIT said to do that. ILO optimized the Style layers to raise the Bullfrog Logit. And this Bullfrog human was created. This is the pure output of a neural network, the first image generated with the ILO deep generative technique.