BU-BIL (Boston University Biomedical Image Library)

BU-BIL is an image library which includes six datasets that represent three imaging modalities and six object types. Providers of the datasets are instructed to choose images that capture the various environmental conditions and imaging noise that arose in their studies. These experts are asked to then select objects from those images that reflect the natural diversity of shape and appearances that these objects can exhibit. The image subregions containing the identified objects are cropped to create the image library. The outcome was a library with 305 objects from 235 images. Authors verify by visual inspection that the image library includes a variety of object appearances, backgrounds, and properties distinguishing objects from the background.

Paper: How to Collect Segmentations for Biomedical Images? A Benchmark Evaluating the Performance of Experts, Crowdsourced Non-Experts, and Algorithms

Image source: How to Collect Segmentations for Biomedical Images? A Benchmark Evaluating the Performance of Experts, Crowdsourced Non-Experts, and Algorithms