[amelius]

What approach does it use? Does it use image segmentation (e.g. U-net) followed by classic image analysis? Or does it get the counts directly from the network?