UNet

A DeepLearning Architecture for image segmentation.

Description

This repository contains the code necessary to run a UNet based on the Dynamic Unet implementation of fastai. The implementation uses the PyTorch DeepLearning framework. UNet is used for image segmentation (pixel-wise classification). The repository contains all code necessary to preprocess large tif-images, run training and validation, and perform predictions using the trained models.

Getting Started

Dependencies

• GDAL, Pytorch-fast.ai, Scipy ... (see installation)
• Cuda-capable GPU (overview here)
• Anaconda (download here)
• developed on Windows 10

Installation

• clone the Stable UNet repository
• conda create --name UNet python==3.9.6
• conda activate UNet
• cd ../UNet/environment
• pip install -r requirements.txt

Executing program

• set parameters in params.py
• run main_unet.py

Help/Known Issues

• None yet

Info

Authors

• Benjamin Stöckigt
• Malik-Manel Hashim

Version History

• 0.1
  • Initial Release

License

Not licensed

Acknowledgments

Inspiration, code snippets, etc.

• fastai
• fastai documentation
• UNet tutorial by Deep Learning Berlin
• UNet adjustable input-channels tutorial by Navid Panchi
• UNet paper