Fig. 7: Improving the quality of the synthetic biofilm dataset using CycleGAN. Generation of the synthetic datasets of mature biofilms by CycleGAN and training the U-Net model.
From: Deep generative modeling of annotated bacterial biofilm images
a The CycleGAN model allows image-to-image translation. This model is used to transfer the characteristics of real biofilms to synthetic biofilms. With this approach, the quality of images has increased. b Training the CycleGAN for the generation of biofilms from late stages of formation. The model allows visualization of the matrix and cell adhesion. c Application of generated datasets for training the U-Net model for segmentation.
