On Direct Distribution Matching for Adapting Segmentation Networks

Georg Pichler, Jose Dolz, Ismail Ben Ayed, Pablo Piantanida
; Proceedings of the Third Conference on Medical Imaging with Deep Learning, PMLR 121:624-637, 2020.

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 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 imaging (MRI) modalities, our approach achieves significantly better results both in terms of accuracy and stability.

Cite this Paper


BibTeX
@InProceedings{pmlr-v121-pichler20a, title = {On Direct Distribution Matching for Adapting Segmentation Networks}, author = {Pichler, Georg and Dolz, Jose and {Ben Ayed}, Ismail and Piantanida, Pablo}, pages = {624--637}, year = {2020}, editor = {Tal Arbel and Ismail Ben Ayed and Marleen de Bruijne and Maxime Descoteaux and Herve Lombaert and Christopher Pal}, volume = {121}, series = {Proceedings of Machine Learning Research}, address = {Montreal, QC, Canada}, month = {06--08 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v121/pichler20a/pichler20a.pdf}, url = {http://proceedings.mlr.press/v121/pichler20a.html}, 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 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 imaging (MRI) modalities, our approach achieves significantly better results both in terms of accuracy and stability.} }
Endnote
%0 Conference Paper %T On Direct Distribution Matching for Adapting Segmentation Networks %A Georg Pichler %A Jose Dolz %A Ismail Ben Ayed %A Pablo Piantanida %B Proceedings of the Third Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2020 %E Tal Arbel %E Ismail Ben Ayed %E Marleen de Bruijne %E Maxime Descoteaux %E Herve Lombaert %E Christopher Pal %F pmlr-v121-pichler20a %I PMLR %J Proceedings of Machine Learning Research %P 624--637 %U http://proceedings.mlr.press %V 121 %W PMLR %X 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 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 imaging (MRI) modalities, our approach achieves significantly better results both in terms of accuracy and stability.
APA
Pichler, G., Dolz, J., Ben Ayed, I. & Piantanida, P.. (2020). On Direct Distribution Matching for Adapting Segmentation Networks. Proceedings of the Third Conference on Medical Imaging with Deep Learning, in PMLR 121:624-637

Related Material