Paper
5 July 2024 The further fully convolutional networks for medical image segmentation
Fengqiao Shen, Yixiang Xu
Author Affiliations +
Proceedings Volume 13184, Third International Conference on Electronic Information Engineering and Data Processing (EIEDP 2024); 131844E (2024) https://doi.org/10.1117/12.3032811
Event: 3rd International Conference on Electronic Information Engineering and Data Processing (EIEDP 2024), 2024, Kuala Lumpur, Malaysia
Abstract
Jonathan Long et al.'s Fully Convolutional Networks (FCN), introduced at CVPR 2015, have been widely adopted for semantic segmentation tasks. In traditional Convolutional Neural Networks, spatial resolution of the input image is gradually reduced through a series of convolution and pooling layers, culminating in a fixed-size feature map, followed by fully connected layers. However, FCN dispenses with the fully connected layers of CNNs and replaces them with fully convolutional layers, enabling pixel-wise mapping between input images and output feature maps. This is achieved by employing transpose convolution (also known as deconvolution) for upsampling the feature maps. Nonetheless, FCN still retains pooling layers. In this paper, we propose improvements to FCN by replacing pooling layers with convolutional layers, and we conduct a series of experiments based on the idea of whether convolutional layers can fully replace pooling layers.
(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.
Fengqiao Shen and Yixiang Xu "The further fully convolutional networks for medical image segmentation", Proc. SPIE 13184, Third International Conference on Electronic Information Engineering and Data Processing (EIEDP 2024), 131844E (5 July 2024); https://doi.org/10.1117/12.3032811
Advertisement
Advertisement
RIGHTS & PERMISSIONS
Get copyright permission  Get copyright permission on Copyright Marketplace
KEYWORDS
Image segmentation

Semantics

Network architectures

Convolution

Image classification

Education and training

Medical imaging

Back to Top