On Direct Distribution Matching for Adapting Segmentation Networks

TL;DR

This paper introduces a direct kernel density matching loss for domain adaptation in segmentation networks, simplifying the process and improving performance over adversarial methods, especially in MRI brain segmentation tasks.

Contribution

It proposes a novel loss function for direct distribution matching in segmentation, eliminating the need for adversarial training and enhancing adaptation quality and stability.

Findings

Outperforms adversarial methods in MRI brain segmentation.

Achieves higher accuracy and stability in domain adaptation.

Simplifies training by unifying distribution matching and segmentation.

Abstract

Minimization of distribution matching losses is a principled approach to domain adaptation in the context of image classification. However, it is largely overlooked in adapting segmentation networks, which is currently dominated by adversarial models. We propose a class of loss functions, which encourage direct kernel density matching in the network-output space, up to some geometric transformations computed from unlabeled inputs. Rather than using an intermediate domain discriminator, our direct approach unifies distribution matching and segmentation in a single loss. Therefore, it simplifies segmentation adaptation by avoiding extra adversarial steps, while improving both the quality, stability and efficiency of training. We juxtapose our approach to state-of-the-art segmentation adaptation via adversarial training in the network-output space. In the challenging task of adapting brain segmentation across different magnetic resonance images (MRI) modalities, our approach achieves significantly better results both in terms of accuracy and stability.