HistoGAN: Controlling Colors of GAN-Generated and Real Images via Color Histograms

Daniel Chodusov Flickr-CC BY-ND 2.0 Target colors GAN-generated images based on specified histogram feature Input image Auto-recolored input image without the need to manually specify target histogram Gertrud K. Flickr-CC BY-NC-SA 2.0 Carl Dunn Flickr-CC BY-NC-SA 2.0 Histograms Figure 1: HistoGAN is a generative adversarial network (GAN) that learns to manipulate image colors based on histogram features. Top: GAN-generated images with color distributions controlled via target histogram features (left column). Bottom: Results of ReHistoGAN, an extension of HistoGAN to recolor real images, using sampled target histograms.