Abstract:   Our knowledge of the brain, the most efficient computing device, is largely limited by our lack of knowledge about the anatomy, including morphology and connectivity, of neurons. Currently, combining serial-section electron microscopy imaging and large scale image processing is the only practical approach for revealing complete neural circuit connectivity. Based on convolutional networks, we build a hybrid-cloud distributed pipeline for large scale image alignment, segmentation, visualization, and analysis. I'll use a zebrafish hindbrain dataset to showcase our pipeline with a focus on convolutional network inference, segmentation, skeletonization, and cell type classification. Combining both morphology and connectivity, we can find some specialized modules in the neural circuits automatically. At last, I’ll give a brief discussion to compare with other approaches, especially flood filling net.