On Improving 3D U-net Architecture

Roman Janovský, David Sedláček, Jiří Žára

2019

Abstract

This paper presents a review of various techniques for improving the performance of neural networks on segmentation task using 3D convolutions and voxel grids – we provide comparison of network with and without max pooling, weighting, masking out the segmentation results, and oversampling results for imbalanced training dataset. We also present changes to 3D U-net architecture that give better results than the standard implementation. Although there are many out-performing architectures using different data input, we show, that although the voxel grids that serve as an input to the 3D U-net, have limits to what they can express, they do not reach their full potential.

Paper Citation

in Harvard Style

Janovský R., Sedláček D. and Žára J. (2019). On Improving 3D U-net Architecture.In Proceedings of the 14th International Conference on Software Technologies - Volume 1: ICSOFT, ISBN 978-989-758-379-7, pages 649-656. DOI: 10.5220/0007830306490656

in Bibtex Style

@conference{icsoft19,
author={Roman Janovský and David Sedláček and Jiří Žára},
title={On Improving 3D U-net Architecture},
booktitle={Proceedings of the 14th International Conference on Software Technologies - Volume 1: ICSOFT,},
year={2019},
pages={649-656},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0007830306490656},
isbn={978-989-758-379-7},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 14th International Conference on Software Technologies - Volume 1: ICSOFT,
TI - On Improving 3D U-net Architecture
SN - 978-989-758-379-7
AU - Janovský R.
AU - Sedláček D.
AU - Žára J.
PY - 2019
SP - 649
EP - 656
DO - 10.5220/0007830306490656