Authors:
Cyril Li
1
;
Christophe Ducottet
1
;
Sylvain Desroziers
2
and
Maxime Moreaud
3
Affiliations:
1
Université de Lyon, UJM-Saint-Etienne, CNRS, IOGS, Laboratoire Hubert Curien UMR5516, F-42023, Saint-Etienne, France
;
2
Manufacture Française des Pneumatiques Michelin, 23 Place des Carmes Déchaux, 63000 Clermont-Ferrand, France
;
3
IFP Energies Nouvelles, Rond-point de L’échangeur de Solaize BP 3, 69360 Solaize, France
Keyword(s):
Neural Network, Electron Tomography, Weakly Annotated Data, U-NET, Contrastive Learning, Semi-Supervised Training.
Abstract:
Segmentation is a notorious tedious task, especially for 3D volume of material obtained via electron tomography. In this paper, we propose a new method for the segmentation of such data with only few partially labeled slices extracted from the volume. This method handles very restricted training data, and particularly less than a slice of the volume. Moreover, unlabeled data also contributes to the segmentation. To achieve this, a combination of self-supervised and contrastive learning methods are used on top of any 2D segmentation backbone. This method has been evaluated on three real electron tomography volumes.