Reproduce “Image Augmentation for Satellite Images” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Oluwadara Adedeji, Peter Owoade, Opeyemi Ajayi, Olayiwola Arowolo

Researchers from Carnegie Mellon University Africa explored the use of Generative Adversarial Networks (GANs) to augment satellite imagery for Land Use and Land Cover (LULC) classification, finding that synthetic data maintained classification performance on a high-accuracy benchmark without substantial degradation.

Key insights

  • GANs can produce visually plausible synthetic satellite images of sufficient quality such that their inclusion in a training dataset does not significantly degrade deep classification model performance on a high-accuracy benchmark.
  • While GAN augmentation did not provide substantial accuracy gains on an already high-performing, geometrically augmented baseline, the consistent performance implies the generated data is of comparable quality to real data.
  • GAN-generated images hold promise for addressing data scarcity and class imbalance in remote sensing by reducing overfitting and enhancing generalizability, particularly in scenarios with genuinely smaller original datasets.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.