High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs

TLDR

Conditional GANs enable many applications, yet their outputs are usually low‑resolution and lack realism. The authors propose a method to synthesize high‑resolution photo‑realistic images from semantic label maps and to generate diverse results for interactive editing. They use conditional GANs with a novel adversarial loss and multi‑scale generator/discriminator to produce 2048×1024 images, incorporate instance segmentation for object manipulation, and enable diverse, interactive edits. Human studies show the method outperforms existing approaches, improving both quality and resolution of image synthesis and editing.

Abstract

We present a new method for synthesizing high-resolution photo-realistic images from semantic label maps using conditional generative adversarial networks (conditional GANs). Conditional GANs have enabled a variety of applications, but the results are often limited to low-resolution and still far from realistic. In this work, we generate 2048 × 1024 visually appealing results with a novel adversarial loss, as well as new multi-scale generator and discriminator architectures. Furthermore, we extend our framework to interactive visual manipulation with two additional features. First, we incorporate object instance segmentation information, which enables object manipulations such as removing/adding objects and changing the object category. Second, we propose a method to generate diverse results given the same input, allowing users to edit the object appearance interactively. Human opinion studies demonstrate that our method significantly outperforms existing methods, advancing both the quality and the resolution of deep image synthesis and editing.