MaybeRichard/zju-eye-pretrain
收藏资源简介:
ZJU Eye-Pretrain数据集是一个统一的多源眼科成像数据集,用于基础模型预训练和下游任务。它包含110万张图像,涵盖26个队列,并采用严格的41列统一清单模式。数据集由三部分组成:私人上海DRI OCT Triton(SS-OCT)图像(419,042张,包含OCT B扫描、彩色眼底和灰度SLO模态)、公共眼底图像(198,629张,主要为彩色眼底图像,部分包含GAMMA OCT)和公共OCT图像(488,705张,为OCT B扫描)。总计1,106,376张图像,覆盖多种眼科疾病和成像技术,如视网膜、OCT、眼底、SLO等,适用于图像分类、分割、生成等任务。数据集还提供每张图像的5个标题(总计550万标题),并支持流式加载以处理大规模数据。许可为混合眼科许可,商业使用受限。
The ZJU Eye-Pretrain dataset is a unified multi-source ophthalmological imaging dataset for foundation model pretraining and downstream tasks. It contains 1.1 million images spanning 26 cohorts with a strict 41-column unified manifest schema. The dataset is composed of three main parts: Private Shanghai DRI OCT Triton (SS-OCT) images (419,042 images, including OCT B-scan, color fundus, and grayscale SLO modalities), public fundus images (198,629 images, primarily color fundus with some GAMMA OCT), and public OCT images (488,705 images, consisting of OCT B-scans). In total, there are 1,106,376 images covering various eye conditions and imaging technologies such as retina, OCT, fundus, and SLO, suitable for tasks like image classification, segmentation, and generation. The dataset also provides five captions per image (totaling 5.5 million captions) and supports streaming for large-scale training. Licensing is mixed under ophthalmology terms with commercial use restrictions.




