Photo retouching is a time-consuming and challenging task that requires advanced skills beyond the abilities of casual photographers. This project aims to construct an agent which learns to edit photo in human-like fashion. Recently, theres has been lots of tasks enhance their robustness and performance via adversarial learning. We borrow the concept of adversarial learning into photo aesthetic improving task, and use the policy gradient to optimize the policy network whose action space is human-like.
chingyaoc/photo-editing-tensorflow: Photo Optimizing Adversarial Net with Policy Gradient Method