Demonstration set of CTL-GAN-generated synthetic fluorescein corneal images for dry eye disease grading (Oxford scale, 100 images per grade)

This demonstration dataset contains de-identified synthetic anterior segment (fluorescein-stained corneal) images for dry eye disease grading on the Oxford scale (Grade 0–5), generated using the collaborative transfer learning with generative adversarial networks (CTL-GAN) framework. It comprises two sets corresponding to the two source institutions in the study, Samsung Medical Center (SMC) and Sun Yat-sen University (SYSU). SMC dataset: synthetic images generated by a StyleGAN3 generator trained directly on the internal SMC cohort of real fluorescein-stained images. SYSU dataset: synthetic images generated from SYSU data using CTL-GAN. The generator weights by the SMC dataset were transferred and adapted to the small SYSU dataset without exchanging any raw images, simulating an institution with limited data benefiting from a pretrained generator. Each set provides 100 synthetic images per Oxford grade, organized into one folder per source and grade. This is a representative subset provided for demonstration and reproducibility, not the full set used for model training. All images are machine-generated; no real patient images are included. Full methodology and evaluation are described in the associated publication.