Fig. 4. Segmentation accuracy of StarDist models across experimental imaging resolutions.

DAPI-stained U2OS nuclei imaged at 10×, 20×, 40× and 60× objectives, with StarDist models trained per resolution. a Segmentation outputs from resolution-specific StarDist models. b Mean ± s.d. Intersection over Union (IoU) for native-resolution models versus those trained on data rescaled to 20×-equivalent. c StarDist predictions on raw 60× data overlaid with Ground Truth. d Side-byside comparison of predictions from 20× and 60×-trained StarDist models.