Edlund et al. created a dataset of phase-contrast images encompassing including 8 morphologically different cell types and trained a CNN-based model to evaluate the model’s cell segmentation accuracy.