Seung Lab develops, maintains, and uses a variety of open-source software. Below is a partial list of the softwares with their corresponding links.
cnpkg (Srini Turaga) – A package for Cortical Network Simulator by Jim Mutch, which is a framework for the fast simulation of cortically-organized networks on CUDA-capable NVIDIA GPUs. cnpkg enables the creation of 3D convolutional networks and also trains them via the backpropagation algorithm. cnpkg was used to train the convolutional networks used in our published retinal reconstructions (Helmstaedter et al., 2013 and Kim et al., 2014). It has been superseded by ZNN.
jQuery DLay plugin (Mark Richardson) – A jQuery plugin used in EyeWire that adds XHR chaining ability. Provides chaining wrappers for jQuery’s AJAX methods and for XHR request, serializing every request in the same queue. Progress and load events are available with reports for active request progress and total queue size. Supports serial or parallel queues.
Trainable Weka Segmentation (Ignacio Arganda-Carreras) – A Fiji plugin that combines a collection of machine learning algorithms with a set of selected image features to produce pixel-based segmentations using Weka. Provides a framework to use and compare any available classifier to perform image segmentation based on pixel classification, bridging machine learning and image processing. Has GUI support.
Watershed (Aleksandar Zlateski) – Modified watershed segmentation used in EyeWire.
ZNN (Aleksandar Zlateski) – Multi-CPU core implementation of convolutional networks in 2D and 3D, including FFT and direct convolution, as well as max filtering.
VAST (Daniel Berger) – Volume Annotation and Segmentation Tool for manual annotation of large EM stacks.
Cloud-Volume – Python client for reading/writing Neuroglancer volumes in the cloud.
Igneous-Scalable Neuroglancer visualization pipeline. Meshing, downsampling, contrast correction, more.
Euclidean Distance Transform – Fast 3D euclidean distance transform for Python that handles anisotropy and multi-labeled images.
Dijkstra’s 3D– Dijkstra’s algorithm for 3D images using 26-connected neighbor vertices.
Connected Components 3D– Connected components for 3D images that supports multiple labels in a single run.
fpzip– Python wrapper for fpzip, a 3D and 4D floating point data lossless compression algorithm