CytoImageNet

CytoImageNet是由多伦多大学创建的一个大规模预训练数据集，包含890,737张来自40个公开可用数据集的显微镜图像，分为894个类别。该数据集通过整合来自多个数据库的图像和标签，模仿了ImageNet的多样性和复杂性。创建过程中，采用了弱标签分配和分层下采样技术，以及图像标准化和上采样处理。CytoImageNet主要应用于生物图像的转移学习，旨在提取生物学上有意义的信息，解决显微镜图像分类任务中的挑战。

CytoImageNet is a large-scale pre-trained dataset developed by the University of Toronto. It contains 890,737 microscopic images sourced from 40 publicly available datasets, and is classified into 894 distinct categories. This dataset replicates the diversity and complexity of ImageNet by integrating images and their corresponding labels from multiple databases. During its development, techniques including weak label assignment, hierarchical downsampling, image normalization and upsampling were employed. CytoImageNet is primarily used for transfer learning on biological images, aiming to extract biologically meaningful information and tackle challenges in microscopic image classification tasks.