Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OpenSR-SRGAN: A Flexible Super-Resolution Framework for Multispectral Earth Observation Data

Simon Donike, Cesar Aybar, Julio Contreras, Luis Gómez-Chova

TL;DR

OpenSR-SRGAN presents a configurable, open-source framework for GAN-based single-image super-resolution tailored to multispectral Earth Observation data. Implemented in PyTorch/PyTorch Lightning, it unifies diverse SRGAN-style architectures with a configuration-driven workflow, enabling flexible experiments without changing code. Key contributions include modular generator/discriminator variants, stabilization techniques (pretraining, ramp-up, EMA), multispectral band support, and ecosystem integration for datasets and evaluation. This framework lowers entry barriers for researchers and practitioners to train, benchmark, and deploy SR pipelines across EO sensors and band configurations.

Abstract

We present OpenSR-SRGAN, an open and modular framework for single-image super-resolution in Earth Observation. The software provides a unified implementation of SRGAN-style models that is easy to configure, extend, and apply to multispectral satellite data such as Sentinel-2. Instead of requiring users to modify model code, OpenSR-SRGAN exposes generators, discriminators, loss functions, and training schedules through concise configuration files, making it straightforward to switch between architectures, scale factors, and band setups. The framework is designed as a practical tool and benchmark implementation rather than a state-of-the-art model. It ships with ready-to-use configurations for common remote sensing scenarios, sensible default settings for adversarial training, and built-in hooks for logging, validation, and large-scene inference. By turning GAN-based super-resolution into a configuration-driven workflow, OpenSR-SRGAN lowers the entry barrier for researchers and practitioners who wish to experiment with SRGANs, compare models in a reproducible way, and deploy super-resolution pipelines across diverse Earth-observation datasets.

OpenSR-SRGAN: A Flexible Super-Resolution Framework for Multispectral Earth Observation Data

TL;DR

OpenSR-SRGAN presents a configurable, open-source framework for GAN-based single-image super-resolution tailored to multispectral Earth Observation data. Implemented in PyTorch/PyTorch Lightning, it unifies diverse SRGAN-style architectures with a configuration-driven workflow, enabling flexible experiments without changing code. Key contributions include modular generator/discriminator variants, stabilization techniques (pretraining, ramp-up, EMA), multispectral band support, and ecosystem integration for datasets and evaluation. This framework lowers entry barriers for researchers and practitioners to train, benchmark, and deploy SR pipelines across EO sensors and band configurations.

Abstract

We present OpenSR-SRGAN, an open and modular framework for single-image super-resolution in Earth Observation. The software provides a unified implementation of SRGAN-style models that is easy to configure, extend, and apply to multispectral satellite data such as Sentinel-2. Instead of requiring users to modify model code, OpenSR-SRGAN exposes generators, discriminators, loss functions, and training schedules through concise configuration files, making it straightforward to switch between architectures, scale factors, and band setups. The framework is designed as a practical tool and benchmark implementation rather than a state-of-the-art model. It ships with ready-to-use configurations for common remote sensing scenarios, sensible default settings for adversarial training, and built-in hooks for logging, validation, and large-scene inference. By turning GAN-based super-resolution into a configuration-driven workflow, OpenSR-SRGAN lowers the entry barrier for researchers and practitioners who wish to experiment with SRGANs, compare models in a reproducible way, and deploy super-resolution pipelines across diverse Earth-observation datasets.

Paper Structure

This paper contains 17 sections, 5 equations, 3 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Statement of Need
  3. Contribution Summary
  4. Software Overview and Framework Design
  5. Generator Architectures
  6. Conditional Generator with Noise Injection
  7. Discriminator Architectures
  8. Training Features
  9. General Training Optimizations
  10. Exponential Moving Average (EMA) Stabilisation
  11. Loss Functions
  12. Limitations
  13. Architecture and Training Components
  14. Internal Metrics
  15. Experimental Configuration and Quantitative Results
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Sentinel-2 Super-Resolution examples: 4$\times$ RGB (top two rows), 8$\times$ SWIR (bottom two rows)
  • Figure 2: False-color visual comparison for $4\times$ RGB super-resolution on SEN2NAIP (Sentinel-2 input $\rightarrow$ NAIP target). Left to right: LR input, model output, HR reference.
  • Figure 3: Visual comparison for $8\times$ multispectral super-resolution (Sentinel-2 6-band input). Left to right: LR input, model output, HR reference. Three bands are shown for visualisation.