Fig. 8

Example segmentations from Segment Anything and Cellpose. A Segment Anything provides context-free segmentations using user prompts to drive the segmentation process; however, we find that using the different modes for SAM, we get poor results for our specific datasets. (i) shows the segmentation using the box-prompt mode, and (ii) shows the segmentation using the point prompt mode. The representative example (ii) shows the model unable to differentiate between a rod and a ring object. (iii) shows the “Whole picture” approach, which fails to single out individual objects in the image. B The original image (i) and predicted outcomes (ii) using Cellpose show an inability to segment rods and rings correctly. Scale bars represent 30 μm