FIQS Dataset (Fundus Image Quality Scores)

We establish a new dataset, Fundus Image Quality Scores (FIQS), including 2246 fundus images with continuous numerical mean opinion scores (MOSs) ranging from 0 to 100 and three-level labels (``Good", ``Usable", and ``Reject"). This dataset was annotated manually by six professional ophthalmologists to represent the real critical factors affecting the fundus image quality. The images in the FIQS dataset have two resolutions: original size (preprocessed image sized around $1942 \times 1942$) and $1024 \times 1024$ (standardized). All opinion scores of each doctor are also included in this dataset to ensure better "rawness".

本研究构建了一款全新的眼底图像质量评分数据集（Fundus Image Quality Scores, FIQS），包含2246张眼底图像，附带取值范围为0至100的连续型数值平均主观评分（Mean Opinion Scores, MOS），以及三级分类标签：“优质”、“可用”与“拒收”。该数据集由六位专业眼科医师手动标注，贴合影响眼底图像质量的真实关键因素。FIQS数据集内的图像提供两种分辨率规格：原始尺寸（预处理后图像尺寸约为1942×1942）与标准化后的1024×1024。为保障数据的原生真实性，本数据集同时收录了每位医师的单独主观评分。