Images created by (x, y) -> (r, g, b) neural networks trained to maximize features from ImageNet-trained CNNs.

This gallery uses a 9-layer tanh -> sigmoid network, using InceptionV3 as the CNN.