papers AI Learner
The Github is limit! Click to go to the new site.

Projection-Based 2.5D U-net Architecture for Fast Volumetric Segmentation

2019-02-01
Christoph Angermann, Markus Haltmeier, Ruth Steiger, Sergiy Pereverzyev Jr, Elke Gizewski

Abstract

Convolutional neural networks are state-of-the-art for various segmentation tasks. While for 2D images these networks are also computationally efficient, 3D convolutions have huge storage requirements and require long training time. To overcome this issue, we introduce a network structure for volumetric data without 3D convolution layers. The main idea is to integrate projection layers to transform the volumetric data to a sequence of images, where each image contains information of the full data. We then apply 2D convolutions to the projection images followed by lifting to a volumetric data. The proposed network structure can be trained in much less time than any 3D-network and still shows accurate performance for a sparse binary segmentation task.

Abstract (translated by Google)
URL

http://arxiv.org/abs/1902.00347

PDF

http://arxiv.org/pdf/1902.00347


Similar Posts

Comments