BasicUNET

Python implementation of a basic U-Net for semantic segmentation of images.

Currently implemented in Tensorflow/Keras, planning to add PyTorch implementation as well.

The following conda command create a known working environment for running the code.

#(Windows Native)
conda create -n unet_test python=3.9 numpy scikit-image matplotlib tqdm cudatoolkit=11.2 cudnn=8.1.0 -c conda-forge
pip install --upgrade pip
pip install "tensorflow<2.11"

#(MacOS - NO GPU)
conda create -n unet_test python=3.9 numpy scikit-image matplotlib tqdm -c conda-forge
pip install --upgrade pip
pip install tensorflow

#(Linux or WSL 2)
#See https://www.tensorflow.org/install/pip#linux_1

Description of scripts

• basic_train_test.py Will train and test a U-Net model. User inputs are outlined in the top of the file under imports.
• performance_metrics.py Holds various performance metrics to evaluate the model.
• support_functions.py Support functions for running the model.
• loss_functions.py Loss functinos that can be used during training.
• view_training_loss.py Function to plot the training loss.