Appion

It is probably related to my fix for #4904.

If this is the case then, you should only see problems on micrographs with weird dimensions. True 4k and 8k images would not have a centering problem.

Our images are true 8K images, 8192 x 8192

Hmm, then the image size should not be changed by the script.

Is this happening only from FindEM or other particle pickers (DoG) as well?

Just FindEM. DoG is fine.

I was looking into the same issue, although in our case the images are from a K2. Removing the fix for #4904 solves the problem for us, but is probably not a good long-term solution.

The plot thickens. Here is a line from a template correlator run:

trimming image from 1024x1024 to 1020x1020 for findem

We've been seeing this issue for ages, always figured it had something to do with FindEM and K2 images.

For what it is worth, FindEM2 does not seem to have any problems with large prime factors and works fine with our K2 images without any additional cropping.

Sorry to hijack this one more time. A fix for those people with odd image dimensions would be to add a new function to appion/appionlib/apImage/imagefilter.py like:

def frame_cut_border(a, newshape):
        mindimx = int(0)
        maxdimx = int(newshape[0])
        mindimy = int(0)
        maxdimy = int(newshape[1])
        return a[mindimx:maxdimx, mindimy:maxdimy]

and substitute this in appion/appionlib/apTemplateCorrelator.py at line 183

                                filtarray = apImage.imagefilter.frame_cut_border(filtarray, newshape)

This will remove some pixels at the bottom right of the image used for the correlation, which will likely not be boxed out in any case, but would preserve the coordinates/centers of the boxed particles from the cropped images.

Can not test it now. Maybe I will do this during Hackathon

No need to test, I just wanted to see what you thought about the code. It's sort of brute force, but I think it's the right way.