Utilities-for-The-Frugal-Labeler

This repository bundles the implementations to the methods used in Methods for the frugal labeler: Multi-class semantic segmentation on heterogeneous labels.

Structure

Network_Training.ipynb

Exemplary notebook which uses the tools provided in this repository.

pymodules/heterogeneous_mask_iterator.py

Implementation of the Heteregenous Mask iterator introduced in the paper. Fully configurable and usable with TensorFlow/Keras.

pymodules/adaptive_objective_functions.py

Implementation of the loss functions introduced in the paper.

pymodules/unet_model.py:

Implementation of the UNet model used in the paper.

pymodules/data.py

Module which provides training- and test-generators ready to be used with Keras.

pymodules/create_one_hot_encoded_map_from_mask.py

Helper module which creates one hot encoded tensors from images (used for training).

Installation

1. Run pip install -e . while in the base directory.

Contact

For questions and remarks feel free to contact Mark Schutera or Luca Rettenberger.