How Automated Segmentation Learns—and Where Humans Help

Describe how an automated segmentation model converts electron-microscopy imagery into candidate neuron fragments.

Distinguish model training, segmentation inference, agglomeration, human proofreading, and connectome materialization.

Explain why image ambiguity can produce false-split and false-merge segmentation errors.

Distinguish directly training a segmentation model from correcting and curating its resulting reconstruction in FlyWire.

Create a small labeled image set for a lightweight segmentation experiment.

Evaluate the baseline segmentation with intersection-over-union on held-out images.

Relate the baseline model’s observed errors to proofreading decisions encountered in FlyWire.