| title | ReFlowSET: Representation-Aligned Latent Flow Matching for SAR-to-EO Image Translation |
|---|---|
| authors | Jeonghyeok Do, Seungchul Lee, Munchurl Kim |
| date | 2026-09-01 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2609.00968 |
| affiliation | KAIST VIClab (Prof. Munchurl Kim) |
| upvotes | 2 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf code |
| abstract | SAR-to-EO image translation aims to generate electro-optical (EO) imagery from synthetic aperture radar (SAR) observations. Existing latent diffusion approaches typically inherit a predetermined autoencoder, although reconstruction fidelity can vary substantially across codecs and modalities. Because the latent codec affects the round-trip preservation of both SAR conditions and EO targets, codec selection constitutes a fundamental design choice; nevertheless, existing methods largely rely on codecs pretrained on natural images. To remedy this, we introduce ReFlowSET, a conditional latent flow-matching framework that selects its codec through a joint SAR--EO reconstruction audit. Rather than inheriting a heavyweight pretrained generator, ReFlowSET trains a substantially smaller conditional DiT from scratch in the selected latent space, using dual-stream SAR conditioning followed by joint feature refinement. To provide semantic guidance for this from-scratch training, intermediate noisy-EO features are aligned with clean target-EO representations extracted by a frozen vision foundation model. This alignment is used only during training and introduces no additional inference cost. Experiments on QXS-SAROPT and SAR2Opt demonstrate state-of-the-art performance across diverse perceptual fidelity and distributional metrics. Code and pretrained weights are publicly available at https://github.com/KAIST-VICLab/ReFlowSET. |
cite
@misc{do2026reflowset,
title={ReFlowSET: Representation-Aligned Latent Flow Matching for SAR-to-EO Image Translation},
author={Jeonghyeok Do and Seungchul Lee and Munchurl Kim},
year={2026},
eprint={2609.00968},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2609.00968}
}