Synthetic Iris Images

This data set was generated using the stable-diffusion-xl-base-1.0 model. The model was fine-tuned with Dreambooth on a private collection of real iris photographs (30 participants, 119 images) without fine-tuning the text encoders. Fine-tuning parameters included the AdamW optimizer, a learning rate of 1e-4, and 10 epochs, with no prior preservation. The prompt used during generation was: "A photo of a [special token] iris."Following synthesis, all generated images underwent a semi-automated re-identification risk assessment using the Cone of Privacy score alongside manual image comparisons to ensure anonymity.This entire dataset consists exclusively of synthetic images. The number in each subdirectory’s name indicates the guidance scale used during image generation.