GaNDLF

GaNDLF, the Generally Nuanced Deep Learning Framework, is an open source framework from MLCommons for building reproducible segmentation and classification models, with a focus on medical imaging and clinical workflows. It exists so that researchers do not have to assemble a training pipeline from scratch. A project is defined by a configuration file describing the intended pipeline and CSV files describing the training data, and the framework handles the rest of the training and inference loop. The framework targets medical imaging researchers, clinical machine learning teams and anyone who needs results that another group can reproduce. It supports multiple deep learning architectures, multiple channels and modalities, and multiple prediction classes in a single pipeline. Data augmentation comes from TorchIO and Albumentations, cross validation is built in with support for parallel computing on high performance clusters, and training can use multiple GPUs on the same machine with automatic mixed precision. The stated design goal is zero code to train a robust model and only low code work for customization, with documentation covering installation, usage, a migration guide, customization of training and inference, and extending the framework itself. Pricing does not apply. GaNDLF is open source, published under the MLCommons organization, with documentation and source freely available and support offered through public GitHub discussions and email rather than a commercial support plan. The framework is described in a 2023 paper in Communications Engineering, which the documentation asks users to cite, making it a reasonable choice for work that will be published.