A Topological Loss Function for Deep-Learning based Image Segmentation using Persistent Homology

This is an unofficial implementation of the TPAMI'20 paper A Topological Loss Function for Deep-Learning based Image Segmentation using Persistent Homology.

In our implementation, we extend the topological loss to multiclass segmentation. The overall loss is the average topological loss over the categories of interest.

Dependencies

• Pytorch 1.0+
• NumPy
• SciPy
• topologylayer

Acknowledgements

This implementation is based on the official implementation of this paper.