SegEM 3D EM segmentation challenge
SegEM challenge training and test data
You can download the volume training data, skeleton training and skeleton test data from the SegEM data repository. It is located in the subfolder SegEM_challenge/. Each file contains one training region (raw data & ground truth) from the cortex dataset. Ground truth and raw data cubes are center-aligned; the raw data is larger to allow for classifiers with large field of view. For details see the Methods of the SegEM paper.
Submission to the SegEM challenge
If you want to participate in the challenge and be evaluated for the results table, please submit your code for processing raw data into segmentations to this e-mail adress: segEMchallenge@brain.mpg.de. The code will then be executed on a non-published test region of the cortex dataset for which dense skeletonization is available. See the Methods of the SegEM paper for details.
Results table
| pixel-based metrics | skeleton-based metrics (published skeletons) | skeleton-based metrics (unpublished skeletons) | |||||||||
| submission name | submission source | submission date | type of method | publication | pixel | warp | rand | var.inf. | node threshold 1 | node threshold 2 | |
| SegEM | Department of Connectomics, MPI BR, Frankfurt | 9/23/2015 | 3D CNN + watershed | Berning et al, Neuron, 2015 [1] | [0.6 - 1.2] E-1 | [3.1 - 9.4] E-4 | [0.1 - 1.3] E-1 | [0.9 - 2.6] E-1 | 1.58 | 4.93 |
[1] = Berning, Boergens, Helmstaedter, SegEM: efficient image analysis for high-resolution connectomics, Neuron, 2015, see 10.1016/j.neuron.2015.09.003