BioImageLoader: Easy Handling of Bioimage Datasets for Machine Learning
This tool addresses data management issues for researchers in bioimaging, but it is incremental as it builds on existing libraries without introducing new algorithms.
The authors tackled the challenge of handling diverse bioimage datasets for machine learning by developing BioImageLoader, a Python library that unifies interfaces for tasks like concatenation and augmentation, enabling applications such as retraining deep learning models and cross-dataset evaluation.
BioImageLoader (BIL) is a python library that handles bioimage datasets for machine learning applications, easing simple workflows and enabling complex ones. BIL attempts to wrap the numerous and varied bioimages datasets in unified interfaces, to easily concatenate, perform image augmentation, and batch-load them. By acting at a per experimental dataset level, it enables both a high level of customization and a comparison across experiments. Here we present the library and show some application it enables, including retraining published deep learning architectures and evaluating their versatility in a leave-one-dataset-out fashion.