Graph neural networks for solving the multicut problem
This repository is a developing of the idea from the paper
Jung, S. and Keuper, M., 2022. Learning to solve minimum cost multicuts efficiently using edge-weighted graph convolutional neural networks.
Proposed changes as of now:
- Unsupervised formulation (including self-prior)
- Relaxed cycle consistency loss
- Learning orthogonal embedding [soon]
More details are available in the slides in the report branch.