NeurIPS 2020

Disentangling Human Error from Ground Truth in Segmentation of Medical Images

Meta Review

The paper proposes an approach to model annotator errors to come up with a consensus annotation from multiple ones. It is essentially an extension of [28] from classification to segmentation. The authors claim this is an extensive change in setting, and the reviewers in the main agree that it is sufficient. The reviewers unanimously agree that the paper is at least marginally above the acceptance threshold.