Fig 3. Evaluation of VolPy performance.
(A) Example of VolPy segmentation results against three manually annotated datasets (mouse sensory cortex left, larval zebrafish tegmental area center, and mouse hippocampus right). Ground truth was built from three different annotations. Matched and mismatched neurons between VolPy (blue) and ground truth (red) were shown in upper and bottom panels respectively. (B) F1 score of VolPy for all evaluated datasets in validation. (C) Average F1 score on training and validation sets grouped by dataset type. Results were provided for training, validation and human annotators (against consensus ground truth) (D) Performance of the network in function of training set size for each dataset type.