Mask Conditional Synthetic Satellite Imagery

TL;DR

This paper introduces a mask-conditional synthetic satellite imagery generation method that enhances land cover classification by augmenting real datasets with diverse synthetic images, achieving performance comparable to real data training.

Contribution

The study presents a novel mask-conditional generator for synthetic satellite images and demonstrates its effectiveness for data augmentation and downstream land cover classification tasks.

Findings

Synthetic imagery achieves similar performance to real imagery in land cover classification.

Combining real and synthetic data improves model accuracy over using only real data.

Diversity in generated outputs is crucial for better downstream task performance.

Abstract

In this paper we propose a mask-conditional synthetic image generation model for creating synthetic satellite imagery datasets. Given a dataset of real high-resolution images and accompanying land cover masks, we show that it is possible to train an upstream conditional synthetic imagery generator, use that generator to create synthetic imagery with the land cover masks, then train a downstream model on the synthetic imagery and land cover masks that achieves similar test performance to a model that was trained with the real imagery. Further, we find that incorporating a mixture of real and synthetic imagery acts as a data augmentation method, producing better models than using only real imagery (0.5834 vs. 0.5235 mIoU). Finally, we find that encouraging diversity of outputs in the upstream model is a necessary component for improved downstream task performance. We have released code for reproducing our work on GitHub, see https://github.com/ms-synthetic-satellite-image/synthetic-satellite-imagery .