search

mbickel / DenseInferenceWrapper

A python wrapper for Krähenbühls dense CRF for medical image volumes.
89 stars 37 forks source link

CRF returning all 0's #5

Open shivin101 opened 7 years ago

shivin101 commented 7 years ago

I am trying to run this code on a set of OCT images they have scaled down to [0,1] of W,H,D and a probability mask in range[0,1] of W,H,D,L where L=1, I am getting a resultant mask of all 0's irrespective of the parameters I give to the CRF Thanks, any help would be appreciated

FangliangBai commented 7 years ago

Hi @shivin101 , I had some similar results. For some images, it returned 0s, while it returned 1s for others. Have you solved this problem?

yunghx commented 7 years ago

Hi, I think I also met this problem. If I understand right, the probability mask including a WHDL matrix, where L means L different probability mappings of different class. So I construct the label tensor with this idea, but the results are matrix all zeros. Do you guys solve it?

RFHO-BDSS commented 6 years ago

I'm also facing this issue, @mbickel do you have some test data that can be used to verify our installations are correct?

when running with verbose I get pre premap -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan

[solved] L must be greater than 1, eg. foreground and background channels. @shivin101 I was doing that as well but tried with 1 - labels as the background and everything works

UpCoder commented 6 years ago

same problem here. Do you solve it?

keesh0 commented 5 years ago

I plan to try RFHO-BDSS's suggestion as I am also obtaining all zer0s. I will let you know my results.

keesh0 commented 5 years ago

I am confused what the feature_tensor should be ? Mine is (W, H, D, 1) where voxels = 0 for background, and = 1 for liver.