Jeonghyeok Do's Paper Accepted for IEEE GRSL

Dr. Jeonghyeok Do's paper has been accepted for publication in IEEE Geoscience and Remote Sensing Letters (GRSL). Congrats!

Journal : IEEE Geoscience and Remote Sensing Letters (GRSL)

Title : ReFlowSET: Representation-Aligned Latent Flow Matching for SAR-to-EO Image Translation

Authors : Jeonghyeok Do, Seungchul~Lee, and Munchurl Kim

Project Page: https://kaist-viclab.github.io/ReFlowSET_site/

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. \rev{Codec selection therefore requires checking preservation of both SAR conditions and EO targets.} To remedy this, we introduce ReFlowSET, a conditional latent flow-matching framework that selects its codec through a joint SAR--EO reconstruction audit. \rev{ReFlowSET retains the frozen codec and trains a 509.3M-parameter conditional Diffusion Transformer (DiT) from scratch, using dual-stream SAR conditioning followed by joint feature refinement.} \rev{For semantic guidance, intermediate noisy-EO features are aligned with a frozen vision foundation model's clean-EO representations during training only, without 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.