The code repository for the 2019 CVPR paper Biologically-Constrained Graphs for Global Connectomics Reconstruction [1]. For more information: https://www.rhoana.org/biologicalgraphs.
This code requires the C++ Graph library from Bjoern Andres: http://www.andres.sc/graph.html [2, 3]. This package should not require additional packages or installation as all functions are included in header files.
git clone https://github.com/Rhoana/biologicalgraphs.git .
cd biologicalgraphs
conda create -n biographs_env python=2.7
conda install --file requirements.txt
Change the variable graph_software_dir in algorithms/setup.py to be the parent directory where you installed the Andres graph repository.
cd algorithms
python setup.py build_ext --inplace
cd ../evaluation
python setup.py build_ext --inplace
cd ../graphs/biological
python setup.py build_ext --inplace
cd ../../skeletonization
python setup.py build_ext --inplace
cd ../transforms
python setup.py build_ext --inplace
Add the parent directory to this repository to your PYTHONPATH variable.
Each new dataset needs a meta file named meta/{PREFIX}.meta where {PREFIX} is a unique identifier for the dataset. All functions in this repository require as input this {PREFIX} identifier to find the locations for the requisite datasets (i.e., image, affinities, segmentation, etc.). An example meta file is provided in neuronseg/meta/Kasthuri-test.meta. This file contains all of the necessary dataset references.
This python package assumes a certain directory structure. Call the parent directory {PARENT_DIR}. All input segmentations should reside in {PARENT_DIR}/segmentations. All meta files must be saved in {PARENT_DIR}/meta.
There is an example script to run the complete framework in the neuronseg/scripts folder. This script runs the entire framework on the testing portion of the Kasthuri dataset [4]. The ground truth for this dataset is in neuronseg/golds and our input segmentations are in neuronseg/segmentations. The meta file neuronseg/meta/Kasthuri-test.meta contains the relevant links and can act as a guide for future datasets. The network architectures used for the results in the paper are included in the subdirectories in neuronseg/architectures.
[1] Matejek, B., Haehn, D., Zhu, H., Wei, D., Parag, T. and Pfister, H., 2019. Biologically-Constrained Graphs for Global Connectomics Reconstruction. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 2089-2098).
[2] Keuper, M., Levinkov, E., Bonneel, N., Lavoué, G., Brox, T. and Andres, B., 2015. Efficient decomposition of image and mesh graphs by lifted multicuts. In Proceedings of the IEEE International Conference on Computer Vision (pp. 1751-1759).
[3] Kernighan, B.W. and Lin, S., 1970. An efficient heuristic procedure for partitioning graphs. Bell system technical journal, 49(2), pp.291-307.
[4] Kasthuri, N., Hayworth, K.J., Berger, D.R., Schalek, R.L., Conchello, J.A., Knowles-Barley, S., Lee, D., Vázquez-Reina, A., Kaynig, V., Jones, T.R. and Roberts, M., 2015. Saturated reconstruction of a volume of neocortex. Cell, 162(3), pp.648-661.