CREMI

Results for the neuron segmentation category, averaged over all samples.

Group	Submission	CREMI score	VOI split	VOI merge	ARAND
harris-manor	DBS3DMT	0.834	1.348	0.226	0.445
harris-manor	DNBS2D3D	0.837	1.331	0.281	0.435
harris-manor	DBS2D3D	0.870	1.350	0.250	0.473
harris-manor	MNBS2D3D	1.079	1.727	0.376	0.555
harris-manor	MBS2D3D	1.133	1.910	0.198	0.610
harris-manor	SBS2D3D	1.161	1.980	0.183	0.624
afish1001	OurUnet	2.197	0.698	5.034	0.855
afish1001	ABWUnet	2.199	0.696	5.042	0.856
afish1001	wunet_wj	2.212	0.641	5.157	0.854
afish1001	unet_wj	2.223	0.648	5.178	0.858
afish1001	wunet_ce	2.244	0.829	4.982	0.882
afish1001	unet_ce	2.246	0.831	4.987	0.883
VIDAR	MALAv2	0.286	0.479	0.127	0.135
VIDAR	unet3d	0.534	0.833	0.282	0.256
VIDAR	mala_py3	0.543	0.724	0.374	0.270
VIDAR	mala_py2	0.675	1.365	0.174	0.300
VIDAR	mala_py	0.696	1.446	0.153	0.307
VCG	mala-seg	0.349	0.452	0.245	0.179
VCG	LFC	0.616	1.085	0.140	0.313
VCG	LearnC	0.618	1.093	0.139	0.313
VCG	LearnCt2	0.623	1.091	0.148	0.316
VCG	testmala	2.412	3.297	3.283	0.888
VCG	ZWZ-0.0	2.675	3.524	4.193	0.927
SeungLab-eding	Seung-ed	1.525	0.895	2.160	0.764
SCI	Submission_1	1.088	1.782	0.556	0.518

CREMI score: The geometric mean of (VOI split + VOI merge) and ARAND.
VOI split and merge: The Variation of Information between a segmentation X and ground truth Y. The split and merge parts correspond to the conditional entropies H(X|Y) and H(Y|X), respectively.
ARAND: The Adapted Rand Error, i.e., 1.0 - Rand F-Score.

For each value, lower is better.

Results for the synaptic cleft detection category, averaged over all samples.

Group	Submission	CREMI score	FP	FN	1 - F-score	ADGT	ADF
VCG	FgDTSm08	434.91	4747.0	355265.0	0.619	42.86	826.96
VCG	FgDT08	795.59	727.0	616222.3	0.834	60.07	1531.10

CREMI score: The mean of ADGT and ADF. We are not including the F-Score for now, as our current way of computing it can lead to unfair comparison. We are working on a more robust measure and will update the results accordingly.
FP, FN, and F-Score: The False Positives, False Negatives, and the resulting F-Score in the clefts detection volume. See metrics for details.
ADGT: The average distance of any found cleft voxel to the closest ground truth cleft voxel.
ADF: The average distance of any ground truth cleft voxel to the closest found cleft voxel.

For each value, lower is better (F-Score shown as 1 - F-Score).

Results for the synaptic partner identification category, averaged over all samples.

Group	Submission	CREMI score	FP	FN
NO2	lr_balan	0.447	175.333	314.000
NO2	lr_affin	0.567	118.333	450.000
PNI	asyn_ori	0.493	310.000	302.333
PNI	asyn_mod	0.423	367.667	262.333
PNI	asyn_at1	0.490	274.333	314.667
PNI	asyn_at0	0.468	293.333	297.000
HCBS	Tr66t-01	0.469	182.667	318.667
HCBS	Tr66comb	0.451	219.000	292.333
HCBS	Tr66_80K	0.449	223.000	286.667
HCBS	PrnTrn66	0.465	178.667	322.333
AnonumoysGroup	PrTrn66	0.620	238.000	451.667
DIVE	PTR_V2	0.453	162.667	333.667
IAL	PSP_unar	0.461	266.667	281.000
IAL	PSP_full	0.464	187.333	310.000

CREMI score: 1 - F-Score of the FP and FN.
FP, FN: The False Positives and False Negatives synaptic partner pairs. See metrics for details.

For each value, lower is better.

MICCAI Challenge on Circuit Reconstruction from Electron Microscopy Images