2 million histopathology stain vectors including Hematoxylin & Eosin color variations

The database includes the stain vectors including color variations of over 2 million patches.

The database is adopted in "Data-driven color augmentation for H&E stained images in computational pathology", (https://www.sciencedirect.com/science/article/pii/S2153353922007830) to check is the color variation of an augmentated sample is acceptable or not. During the training, the color variation of an augmented sample is compared with the variations included in the database. If N variations from the database are found within a radius R from the components of the augmented sample, the augmented sample is considered acceptable (in terms of color variations); otherwise, it is discarded.

The database is stored in a .pickle file, including vectors with six elements: the RGB components of Hematoxylin and Eosin, for every patch. Double entries are removed from the database. Code to import and to extend database with new data is available here: https://github.com/ilmaro8/Data_Driven_Color_Augmentation

Stain vectors are collected from six private and public sources, to ensure that color variations can cover the variability of H&E-stained tissues: TCGA, ExaMode colon dataset, Camelyon, Puerta del Mar, Clinic, CAD.