Delira Baseline: Unet-Segmentation

This repository contains a U-Net implementation which is compatible with thedelira-Framework and itsPyTorch-Backend.

Contents

It contains the following:

• Generic U-Net implementation for 2D and 3D

• Implementation of the SoftDiceLoss

• Implementation of the RADAM Optimizer

• Implementation of Stat-Score Calculation (TP, FP, TN, FN)

• Implementation of a Multiclass-Dice Score

• Utility Functions to convert tensors from index based formats to onehot encoding

• An exemplaric training script

• An exemplaric prediction script

Usage

The easiest usage is to install it, and fill out the missing parts of the example scripts (essentially the data-handling).

Installation

This repository contains an installable python-package structure which can be installed by pip viapip install git+https://github.com/justusschock/delira-unet-baseline

Remaining ToDos:

• Add Unittests for CI/CD

• Add pretrained networks and predefined datasets