Deep Learning training module for semantic segmentation in histological images. The training dataset is the Lizard (https://zenodo.org/record/7508237) dataset. The dataset comprises 4981 patched images from multiple colon tissue H&E stained histological images. Each image contains a segmentation mask with six nuclei classes. The classes are neutrophil epithelial, lymphocyte, plasma, eosinophil, connective tissue. The training can be done both deterministically and non deterministically on three different architectures: A basic U-Net, a context U-Net and a spatial transformer U-Net.
- Free software: MIT
- Documentation: https://seg-training.readthedocs.io.
- Fully reproducible mlf-core Pytorch model
- Allows training of pytorch models with the following structures:
- U-Net
- Context U-Net
- Spatial Transformer U-Net
This package was created with mlf-core using Cookiecutter.