SAR2Earth: A SAR-to-EO Translation Dataset for Remote Sensing Applications
Keywords: AI for social good, Dataset, SAR-to-EO translation, Benchmark
TL;DR: This paper introduces SAR2Earth, a benchmark dataset for translating SAR images into EO-like representations for better remote sensing analysis.
Abstract: Electro-optical (EO) images are essential to a wide range of remote sensing applications. With the advent of data-driven models, the efficiency of EO image analysis has significantly improved, enabling faster and more effective outcomes in these applications. However, EO images have inherent limitations—they cannot penetrate cloud cover and are unable to capture imagery at night. To overcome these challenges, synthetic aperture radar (SAR) images are employed, as they can operate effectively regardless of weather conditions or time of day. Despite this advantage, SAR images come with their own difficulties: they are affected by speckle noise, complicating analysis, and existing algorithms developed for EO imagery are not directly transferable to SAR data. To address these issues, we introduce SAR2Earth, a benchmark dataset specifically designed for SAR-to-EO translation. By translating SAR images into EO-like representations, SAR2Earth allows the extensive range of algorithms developed for EO imagery to be applied effectively to SAR data. The dataset consists of 18 spatially aligned pairs of SAR and EO images, collected from 8 distinct regions encompassing both urban and rural. We provide comprehensive evaluations, detailed model analyses, and extensive experimental results. All codes and datasets will be made publicly available at https://sar2earth.github.io.
Primary Area: datasets and benchmarks
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 9259
Loading