Figure 3
From: Automatic enhancement preprocessing for segmentation of low quality cell images
Overview of AEP+AWEL architecture. When we use a segmentation dataset of three classes, we set three translation filters. Each translation filter is added to an input cell image, and we obtain three translation images. Each translation image emphasizes objects related to the segmentation result. Translated images are fed to the second network one-by-one for segmentation, and we compute the loss for all segmentation images using AWEL.
