Fig. 4: Study of the generated cell images and analysis of various defects for subsequent filtration.
From: Deep generative modeling of annotated bacterial biofilm images
a The most frequent defects in the generated data. b Calculation of the percentage of images with multipart cells. c Distributions of the length-to-width ratio and convexity for cell images from each synthetic dataset. d Calculating the total number of images with artifacts from each generative model. e The implementation of filtering datasets has led to an improvement in the FID metric and quality of the obtained synthetic data.
