Cell Image Segmentation by Integrating Pix2pixs for Each Class
Abstract
This paper presents a cell image segmentation method using Generative Adversarial Network (GAN) with multiple different roles. Pix2pix is a kind of GAN can be used for image segmentation. However, the accuracy is not sufficient because generator predicts multiple classes simultaneously. Thus, we propose to use multiple GANs with different roles. Each generator and discriminator has a specific role such as segmentation of cell membrane or nucleus. Since we assign each generator and discriminator to a different role, they can learn it efficiently. We evaluate the proposed method on the segmentation problem of cell images. The proposed method improved the segmentation accuracy in comparison to conventional pix2pix.
Cite
Text
Tsuda and Hotta. "Cell Image Segmentation by Integrating Pix2pixs for Each Class." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019. doi:10.1109/CVPRW.2019.00139Markdown
[Tsuda and Hotta. "Cell Image Segmentation by Integrating Pix2pixs for Each Class." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019.](https://mlanthology.org/cvprw/2019/tsuda2019cvprw-cell/) doi:10.1109/CVPRW.2019.00139BibTeX
@inproceedings{tsuda2019cvprw-cell,
title = {{Cell Image Segmentation by Integrating Pix2pixs for Each Class}},
author = {Tsuda, Hiroki and Hotta, Kazuhiro},
booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
year = {2019},
pages = {1065-1073},
doi = {10.1109/CVPRW.2019.00139},
url = {https://mlanthology.org/cvprw/2019/tsuda2019cvprw-cell/}
}