NeuralNetworkTools

Guide page ›› Code on Github ››

This toolbox contains the general building blocks for neural networks that are used by the SegmentationTools. It can make (UNET) networks (MakeUnet) and classification networks (MakeClassifyNetwork), and contains loss layers (Dice, Jaccard, Tversky, cross-entropy, focal and top-K), configurable activation layers and functions to encode and decode classes. There are also functions to change the dimensions of a network, to freeze and copy layers for transfer learning, and to inspect network nodes and features (NetSummary, GetNetNodes, AnalyzeNetworkFeatures). Back››