Seamless Trans-X Lab, Yonsei University
Project pages for published research from Yonsei STL. Each link below keeps the existing project URL.
Results in Engineering, 2026
A multispectral multimodal language model and the DENSE-QA benchmark for adverse-weather driving scene understanding, using synchronized RGB and multispectral images.
Information Fusion, 2026
A large-scale driving dataset of 100,000 aligned RGB-SWIR pairs, with benchmarks for object detection and RGB-to-SWIR translation in adverse conditions.
WACV, 2026
A dual-token transformer for infrared super-resolution that combines learned regional priors with local features, evaluated on M3FD, RASMD, and TNO.
Computer Vision and Image Understanding, 2026
A diffusion model that translates RGB images to infrared using semantic segmentation and a Spectral Attention Block, evaluated on FLIR, FMB, MFNet, IDD-AW, and RANUS.
Measurement, 2026
A measurement-driven comparison of RGB and SWIR imaging in real rainy driving scenes using 620 paired frames, 14 image-quality metrics, visibility estimation, and fixed pretrained perception models.
IEEE Access, 2025
MWSC jointly predicts multiple weather conditions and one severity level using global–local and image–text cross-attention, evaluated on 40,205 annotated driving images.
Technologies, 2025
An RGB-to-NIR translation model that uses a vision foundation model, cross-attention, and a multi-scale PatchGAN, with results on RANUS, IDD-AW, and zero-shot BDD100K.
ICCV Workshops, 2025
IVIFormer adapts infrared–visible image fusion to day and night by switching the query and key/value domains in cross-attention, guided by a lightweight illumination classifier.
Applied Sciences, 2024
SWIR-based material discrimination of visually similar objects: real and artificial fruits and vegetables, humans, mannequins, and banners, using moisture-sensitive spectral bands.