CytoImageNet

CytoImageNet is a large-scale pretraining dataset of microscopy images (890K, 894 classes). In the paper, CytoImageNet pretraining yielded features competitive to and different from ImageNet pretrained features on downstream microscopy tasks.

• It was constructed from 40 openly available microscopy datasets.
• Weak labels (from experimental metadata) were assigned to each image in the dataset.
• Images are of varying sizes.

The primary purpose of the dataset is to be used for pretraining as a pretext task for learning useful bioimage representations. However, it may be used for validation or exploratory analysis.