Thank you both, Eduardo and Dahonora, for the help!
I have now located the problem I had with my setup.
It was in the parsing of my data. The way I labelled my images was leading to a wrong sequence when analysing the images with glob.
I labelled my images as 0-1,0-2 … 0-10. And when using glob the 10th image was read before the 2nd… (rookie mistake I guess). But now I recieve results close to what my roughly estimated result and result from other software were.