Deep Snow: Synthesizing Remote Sensing Imagery with Generative Adversarial Nets

TL;DR

This paper demonstrates that GANs can generate realistic remote sensing images with meaningful changes, introduces metrics for evaluating transformation quality, and analyzes artifacts affecting image realism and distinguishability.

Contribution

It presents a method for synthesizing remote sensing imagery using GANs in unpaired settings and introduces deep embedding metrics to assess transformation quality.

Findings

GANs can produce realistic remote sensing images with perceptual similarity.

Deep embedding metrics help visualize training dynamics and image distinguishability.

Artifacts in generated images impact their similarity to real images.

Abstract

In this work we demonstrate that generative adversarial networks (GANs) can be used to generate realistic pervasive changes in remote sensing imagery, even in an unpaired training setting. We investigate some transformation quality metrics based on deep embedding of the generated and real images which enable visualization and understanding of the training dynamics of the GAN, and may provide a useful measure in terms of quantifying how distinguishable the generated images are from real images. We also identify some artifacts introduced by the GAN in the generated images, which are likely to contribute to the differences seen between the real and generated samples in the deep embedding feature space even in cases where the real and generated samples appear perceptually similar.