nnU-Net

nnU-Net is an open-source framework for researchers who need to segment biomedical images using labeled training data. It examines each dataset and chooses suitable U-Net configurations, reducing the amount of pipeline design needed for a new imaging task.

Its main distinction is dataset-specific adaptation. The framework uses the characteristics of the training data to determine preprocessing and network configuration, rather than expecting one fixed model setup to fit every dataset. It covers preprocessing, training, model selection and inference within the same framework.

nnU-Net supports both 2D and 3D images, multiple image formats and custom channel definitions. That makes it relevant to biomedical datasets and non-standard imaging problems where models pretrained on natural images may be a poor fit. Training from scratch is a particular focus.

For researchers developing segmentation methods, it also provides a baseline and an extensible framework for testing their own approaches. Available network choices include residual encoder U-Nets. It only creates a cascaded configuration when the dataset's characteristics justify one.

The software is written in Python, uses PyTorch and carries the Apache 2.0 license. Its models are designed for supervised semantic segmentation, where labeled examples teach the network which image regions belong to each class.