Figure 3
From: Dense cellular segmentation for EM using 2D–3D neural network ensembles
Results. (a, b) Orthoslice of Subject 1 image and segmentation. (c) Test dataset orthoslice, segmented cell highlighted. (d, f) Comparison between ground truth segmentation of test cell and our best 2D and 3D algorithms. (g) Annotator comparison (AC) dataset orthoslice, segmented cell highlighted. (h, j) Annotator comparison cell segmentations, comparing the two human annotators and our best (3D) algorithm. (k) Summarized comparison of mean intersection-over-union across organelle classes (\({\rm{MIoU}}^{(org)}\)) on test and evaluation datasets for segmentation algorithms. For full results, see Table S1. (m) Comparison of organelle volume fractions between two human annotators and our best algorithm, computed from annotator comparison cell segmentations.
