Skip to main content

Table 7 Segmentation performance for the FIB-SEM dataset

From: Effective automated pipeline for 3D reconstruction of synapses based on deep learning

Size of pixel Metrics Becker [15] U-Net [23] F-RCNN + F-RCNN + F-RCNN + fusion +
overlap     Growing [19] GrabCut continuity + GrabCut
0-pixel AP 79.3% 70.3% 66.8% 79.4% 82.8%
  F1 78.2% 68.9% 69.4% 77.5% 82.9%
  Jaccard 61.5% 54.1% 53.5% 65.2% 70.9%
3-pixel AP 88.0% 78.8% 74.0% 88.1% 91.6%
  F1 84.4% 77.2% 74.6% 85.4% 89.4%
  Jaccard 74.6% 60.9% 60.2% 75.7% 79.6%
5-pixel AP 90.2% 80.6% 79.5% 90.8% 93.0%
  F1 86.6% 79.3% 78.9% 87.8% 91.1%
  Jaccard 80.3% 67.5% 67.7% 80.6% 83.7%