Machine Learning GAN Deconvolution

We initially benchmarked our GAN against the LUCYD network using a Z-stack of a selection U2OS cells acquired by widefield microscopy from the Li et al dataset (Li et al., 2022) . Following this we ran another benchmark against Deconwolf to gain a direct comparison of the improvements achieved using a Z-stack image from the ChrX-36plex OligoFISSEQ dataset (Nguyen et al., 2020). Data is organised as follow: ├───Input└───Output ├───Deconwolf ├───GAN └───LUCYD我们首先将生成对抗网络（Generative Adversarial Network，简称GAN）与LUCYD网络开展基准对比测试，所用图像源自Li等人（Li et al., 2022）公开数据集，为经宽场显微镜采集的筛选后U2OS细胞的Z堆叠（Z-stack）图像子集。随后，我们又以Deconwolf为基准对象进行了另一组基准测试，采用ChrX-36plex OligoFISSEQ数据集（Nguyen et al., 2020）中的Z堆叠图像，以直接对比验证本方法所实现的性能改进效果。数据集组织形式如下： ├── 输入（Input） └── 输出（Output） ├── Deconwolf ├── GAN（生成对抗网络） └── LUCYD