neurite

Neurite: A modular deep learning library for medical image processing.

Neurite is organized into two main parts:

• Backend-specific modules: Each backend (e.g., PyTorch, TensorFlow) is contained in their own subpackage with functionalities such as layers and models.
• Core Python utilities: Common functionality implemented in pure Python and NumPy, organized under the 'py' subpackage.

Configuration

The backend is determined by the environment variable NEURITE_BACKEND. The default backend is TensorFlow unless NEURITE_BACKEND is set to 'pytorch'.

neurite/torch is the package of neurite that handles its PyTorch implementation.

MODULE	DESCRIPTION
layers	Operations and augmentations for model training.
losses	Loss functions for various learning objectives.
random	Utilities for random sampling, distributions, and random number generation for augmentations and model training.
utils	Helper functions and utilities for common PyTorch operations, including tensor manipulation.
modules	Foundational, trainable layers for constructing neural networks, including normalization layers and convolutional blocks.
models	Prebuilt yet flexible neural network architectures designed for specific tasks, such as image segmentation, registration, or classification. models leverage layers and modules from other components of the neurite for streamlined object construction.