Tal Shiri, Joel Simon

Train a DCGAN on Perlin Noise to generate shapes, cut the generated shapes from an Artbreeder image, the cutouts and placement are optimized with feature inversion over a face classifier