Gatys et al. (2015) showed that optimizing pixels to match features in a convolutional network with respect reference image features is a way to render images of high visual quality. We show that unrolling this gradient-based optimization yields a recurrent computation that creates images by incrementally adding onto a visual "canvas". We propose a recurrent generative model inspired by this view, and show that it can be trained using adversarial training to generate very good image samples. We also propose a way to quantitatively compare adversarial networks by having the generators and discriminators of these networks compete against each other.
@article{arxiv.1602.05110,
title = {Generating images with recurrent adversarial networks},
author = {Daniel Jiwoong Im and Chris Dongjoo Kim and Hui Jiang and Roland Memisevic},
journal= {arXiv preprint arXiv:1602.05110},
year = {2016}
}