MIPCandy

MIP Candy is an open source Python framework for medical image processing research, built by Project Neura. It provides ready to use training, inference and evaluation pipelines so researchers can prototype models without writing repetitive boilerplate. The tagline describes it as a candy for medical image processing and next generation infrastructure for fast prototyping in machine learning. The framework is aimed at machine learning researchers and students working on tasks such as medical image segmentation. Users start from a SegmentationTrainer class and implement only the network construction, while the library handles data flow, loss computation, augmentation, checkpointing and evaluation. The site highlights informative console logs that track metric trends across epochs and per case validation results, built in overlaid previews of model output, simple 3D volume rendering, automatic metric plotting, and a profiler for finding memory and compute problems at sub epoch level. A dashboard can connect to Notion, Weights & Biases and TensorBoard for remote experiment tracking. The quick start example trains on the ACDC cardiac dataset and is said to replicate most of the features of a widely used segmentation baseline, without augmentations. The site names research groups at the University of Toronto, the Vector Institute and related student organizations as users. Documentation is linked from the home page. The vendor does not publish pricing, and the project is presented as freely available software.