cassetta

This package is a versatile deep-learning toolbox for PyTorch, tailored to researchers working with N-dimensional vision problems, and more specifically medial imaging problems.

It is intended to provide building blocks for a wide variety of architectures, as well as a set of pre-defined backbones, as well as a few task-specific models (segmentation, registration, synthesis, ...).

It will not provide domain-specific tools with dedicated pre- and post- processing pipelines. However, such high-level tools can be implemented using this toolbox.

MODULE	DESCRIPTION
models	Task-specific models.
backbones	Task-agnostic architectures to use as backbones in models.
layers	Building blocks for backbones and models.
losses	Differentiable functions to optimize during training.
metrics	Non-differentiable functions to compute during validation.
training	Tools to train networks.
inference	Tools to apply networks to unseed data.
functional	Lower-level functional utilities.
io	Input/output.
core	Core utilities, mostly intended for internal use.