GeoSET: Generalist Foundation Model for SAR-to-EO Image Translation
GeoSET is pitched as a reusable parent model for SAR-to-EO translation, trained on more than 3 million paired images.
The paper says existing methods are usually tied to one dataset and narrow sensing conditions. GeoSET uses a speckle-robust SAR encoder and a pretrained conditional generator to handle varied sensors, resolutions, and ground sampling distances. It can be adapted to downstream datasets with LoRA, updating 0.60% of generator parameters in about an hour per dataset. The authors report state-of-the-art FID and DISTS results across six benchmarks. HF Daily Papers' note
The paper says existing methods are usually tied to one dataset and narrow sensing conditions. GeoSET uses a speckle-robust SAR encoder and a pretrained conditional generator to handle varied sensors, resolutions, and ground sampling distances. It can be adapted to downstream datasets with LoRA, updating 0.60% of generator parameters in about an hour per dataset. The authors report state-of-the-art FID and DISTS results across six benchmarks. HF Daily Papers' note
score 5