<b>A single-centre ultra-widefield fundus image dataset for multi-disease retinal recognition</b>

All images are 133° frontal F2 true-color fundus images, defined in this project as macula-centered frontal images with the optic disc visible whenever possible.The dataset includes multiple image file formats, including TIFF, JPG, and PNG. Each image is assigned a de-identified image ID and is accompanied by image-level multi-label annotations for retinal disease classification and image quality assessment.The disease label system includes normal, diabetic retinopathy (DR), retinal vein occlusion (RVO), high myopia / pathological myopia-related changes, and other retinal diseases.

AMD or macular lesions are not used as an independent label in this version and may be included under other_disease when applicable. A derived multi_disease field indicates images with two or more positive disease labels.The dataset is designed to support research on multi-label retinal disease classification, image quality assessment, patient-level train/validation/test splitting, and deep learning benchmark evaluation using ultra-widefield fundus images.