Overview and example segmentations. A) Original 1000x1000-pixel EM image of the fruit fly ventral nerve cord. B) Manual ground-truth segmentation. C) The result of our segmentation algorithm after k-means clustering. D-G) Segmentations of the highlighted region in yellow returned by each algorithm using a total of roughly 1000 superpixels each (boundaries shown in red). Our algorithm better adheres to the true edges compared to Watershed, SLIC, and TurboPixels.