Prediction gets stuck

[hamid13r]

Dear all,

I was trying to follow the fidder tutorial for Warptools session but the python code gets stuck forever when trying to predict the fiducial locations. Then I noticed the same thing happens to the command-line usage:
I run this command:
fidder predict --input-image frames/average/JTL016A_2_L01_ts_002_001_0001_-0.0_May04_02.40.07.mrc --probability-threshold 0.5 --output-mask mask.mrc

And I get this output:

fidder_output.txt

And nothing happens for a very long time (hours I believe).

Any idea what I am doing wrong? I even tried to bin the micrograph to make it very small, hoping to make it faster.

Thanks,
Hamid

[alisterburt]

Hi Hamid,

Are you comfortable dropping into the Python layer and running things manually to see where things get stuck? I suspect it's struggling to count pixels in each subregion in a poor segmentation

[hamid13r]

Oh it runs in a second in the python layer, and as you predicted the segmentation was pretty poor (a lot of membrane etc were segmented as fiducial).

Is there anything other than thew probability threshold to change?

[alisterburt]

Can you send a picture of the segmentation? Probability threshold and if that fails adding some of your data to the training data and retraining would be my recommendations

[hamid13r]

This is the image:

and this is the mask

(I just did plt.imshow so that is probably why there is a rotation.)

[alisterburt]

And if you look at the probability threshold is there a threshold that works better?

[hamid13r]

No, seems like the membrane somehow scored pretty high here.
Sorry I should have included this in the last post, this is threshold=0.99

[alisterburt]

looks like it'll need a few extra training images! If you send a few (5? at different tilts) across I'll add them and retrain

[hamid13r]

That would be amazing.

I will email you a dropbox link.