Enhancing Photorealism Enhancement

TL;DR

This paper introduces a novel adversarial training approach with architectural improvements to enhance the realism of synthetic images, addressing dataset biases and achieving significant improvements over existing methods.

Contribution

It proposes a new training strategy and network architecture for photorealism enhancement, improving stability and realism in synthetic image rendering.

Findings

Substantial gains in image realism and stability.

Effective mitigation of dataset bias effects.

Outperforms recent image-to-image translation methods.

Abstract

We present an approach to enhancing the realism of synthetic images. The images are enhanced by a convolutional network that leverages intermediate representations produced by conventional rendering pipelines. The network is trained via a novel adversarial objective, which provides strong supervision at multiple perceptual levels. We analyze scene layout distributions in commonly used datasets and find that they differ in important ways. We hypothesize that this is one of the causes of strong artifacts that can be observed in the results of many prior methods. To address this we propose a new strategy for sampling image patches during training. We also introduce multiple architectural improvements in the deep network modules used for photorealism enhancement. We confirm the benefits of our contributions in controlled experiments and report substantial gains in stability and realism in comparison to recent image-to-image translation methods and a variety of other baselines.