Login | Register

CREMI

MICCAI Challenge on
Circuit Reconstruction from Electron Microscopy Images

Data Format Specification

Volumes and annotations are stored in a single HDF5 file with the following datasets:

Volumes

The volumes are stored in row-major format, i.e., with dimensions (depth,height,width). All volumes have an attribute resolution that specifies the size of voxels (depth,height,width) in nm. In the padded versions of the datasets, the raw volume provides much more context. In this case, the label volumes have an additional offset attribute (z,y,x) in nm that marks the beginning of the label volumes in the raw volume.

volumes/raw
gray scale pixels
volumes/labels/clefts
synaptic cleft segmentation
volumes/labels/neuron_ids
neuron segmentation

Annotations

Annotations for pre- and post-synaptic partner pairs are provided as well. All locations are given as (z,y,x) in nm. In the padded versions of the datasets, the raw volume provides much more context. In this case, the annotation group has an additional offset attribute (z,y,x) in nm, which is the same as for the label volumes. All locations are relative to this offset, such that they are the same as in the unpadded version.

annotations/ids
list of ids
annotations/locations
list of annotation locations as (z,y,x) in nm, same order as ids
annotations/type
list of annotation types (presynaptic_site or postsynaptic_site), same order as ids
presynaptic_site/partners
list of id tuples (pre, post) linking pre- and postsynaptic partners
comments/target_ids
list of ids for annotations that have a comment
comments/comments
list of comments, same order as target_ids

Training Volumes

We provide three training volumes of adult Drosophila melanogaster brain imaged with serial section Transmission Electron Microscopy (ssTEM).

Training data is never perfect. Over the course of the coming weeks, we will keep refining the training data and provide follow up versions in addition to this initial version. Please follow us on Twitter for updates.

Dataset A

Downloads: cropped version (175 MB), padded version (1.47 GB), Browse online

Dataset B

Downloads: cropped version (167 MB), padded version (1.46 GB), Browse online

Dataset C

Downloads: cropped version (173 MB), padded version (1.49 GB), Browse online

Test Volumes

We provide three test volumes, that have similar characteristics to the provided training samples.

Dataset A+

Downloads: cropped version (151 MB), padded version (1.46 GB)

Dataset B+

Downloads: cropped version (152 MB), padded version (1.44 GB)

Dataset C+

Downloads: cropped version (152 MB), padded version (1.47 GB)

Acknowledgements

Image Data
Sample preparation and sectioning by Rick Fetter; imaging by Zhihao Zheng, Cam G. Robinson, Daniel Milkie, Eric Perlman, John Price, Davi Bock; intensity correction by Eric Perlman, Misha Kazhdan; stitching and alignment by Khaled Khairy, Bill Karsh, Eric Trautman, Eric Perlman.
Neuron Segmentation
Initial segmentations by Peter Li and Jan Funke, proof-read and fixed by Ariadne Service GmbH and Chris Ordish, partially reviewed by Stephan Saalfeld and Jan Funke.
Synaptic Cleft Segmentation
Provided by Julia Buhmann and Chong Zhang, reviewed by Jan Funke.
Synaptic Partner Annotations
Provided by Jan Funke, reviewed by Julia Buhmann.