Michał Futrega, Alexandre Milesi, Michal Marcinkiewicz, Pablo Ribalta
TL;DR本文通过一项消融研究，建议了优化的U-Net架构用于BraTS21挑战的脑肿瘤分割任务，并在NVIDIA Deep Learning Examples GitHub仓库中开源了代码来重现我们的BraTS21提交。
Abstract
We propose an optimized u-net architecture for a brain \mbox{tumor} segmentation task in the BraTS21 Challenge. To find the \mbox{optimal} model architecture and learning schedule we ran an extensive ablation study