GeoSET: Generalist Foundation Model for SAR-to-EO Image Translation
GeoSET is a generalist SAR-to-EO translation model pretrained on over three million pairs and adapted with LoRA.
GeoSET is presented as the first generalist foundation model for SAR-to-electro-optical image translation, built as one pretrained parent adapted to downstream datasets. Authors curated more than 3 million high-quality SAR–EO pairs from over 10 million SAR observations spanning sensors, resolutions, and regions. A speckle-robust SAR encoder bridges SAR inputs to a pretrained conditional generator. LoRA updates about 0.60% of generator parameters in roughly one hour per dataset, and GeoSET reports state-of-the-art FID and DISTS on six benchmarks with full fine-tuning or LoRA.
- First generalist parent model for SAR-to-EO image translation.
- Over 3 million SAR–EO pairs from more than 10 million SAR observations.
- Speckle-robust SAR encoder conditions a pretrained image generator.
- LoRA updates 0.60% of parameters in about one hour per dataset.
- State-of-the-art FID and DISTS on six downstream benchmarks.
From the original article
1 min read at huggingface.co
Paired synthetic aperture radar (SAR) and electro-optical (EO) imagery is increasingly available across sensors, resolutions, and geographic regions. Yet existing SAR-to-EO image translation (SET) methods are typically trained on a single, limited-scale dataset, producing models specialized to particular sensing conditions.
By Jeonghyeok Do, Munchurl Kim · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.