EED-Astig

We collected data from 640 children aged 3-6 years who underwent routine vision examinations at the outpatient department of Peking University First Hospital between January 1 and June 30, 2025. We present EED-Astig, a new multimodal pediatric dataset for AI-based astigmatism severity prediction.The dataset is released as a single compressed file named `EED-Astig.zip`, which, upon extraction, reveals two main directories: **Images** and **Annotations**, along with an additional file **Parameter.xls**, each containing 3088 files. The Images folder stores all periocular photographs in JPEG format (.jpg), named according to the convention `[group_index].[view_index].jpg`, where `group_index` identifies the subject group and `view_index` (ranging from 1 to 5) specifies the capture perspective—1 for frontal view of both eyes, 2 and 3 for frontal views of the right and left eye respectively, and 4 and 5 for side views of the right and left eye. The Annotations folder contains a corresponding JSON file for every image, sharing the same filename, with annotation content varying by view type: frontal-view images (.1, .2, .3) are annotated with keypoint coordinates (e.g., canthi, lower eyelid points) and corneal segmentation masks, while side-view images (.4, .5) contain keypoints marking the start and end of the eyelash line, all meticulously labeled by professional ophthalmologists and rigorously verified to ensure high data quality and accuracy. The Parameter.xls file provides demographic and clinical metadata, including gender, height, screen time, and birth history, for each subject group, enabling comprehensive supervised learning for astigmatism severity assessment.