Deep Learning training module for semantic segmentation in histological images.
The training dataset used is the Lizard dataset: https://zenodo.org/record/7508237
The dataset comprises 4,981 patched images from multiple colon tissue H&E-stained histological images. Each image contains a segmentation mask with six nuclei classes:
- Neutrophil
- Epithelial
- Lymphocyte
- Plasma
- Eosinophil
- Connective tissue
Training can be performed both deterministically and non-deterministically on three different architectures:
- Basic U-Net
- Context U-Net
- Spatial Transformer U-Net
- Free software: MIT License
- Documentation: https://seg-training.readthedocs.io
- Fully reproducible mlf-core PyTorch model
- Supports training with the following architectures:
- U-Net
- Context U-Net
- Spatial Transformer U-Net
- Clone the repository:
git clone <repository-url>
cd seg_training- Create the conda environment using the provided
environment_new.ymlfile:
conda env create -f environment_new.yml
conda activate <environment-name>- U-Net: https://link.springer.com/chapter/10.1007/978-3-319-24574-4_28
- Context U-Net: https://link.springer.com/chapter/10.1007/978-3-319-75238-9_25
- mlf-core: https://mlf-core.readthedocs.io/en/latest/
- Cookiecutter: https://github.com/audreyr/cookiecutter
This package was created with mlf-core using Cookiecutter.
This repository was originally written by Dominik Molitor.