Impossible to interact with OpenSfM from Jupyter notebook

Hello,

I'm having issues in accessing the OpenSfM functionality from a jupyter notebook. When accessing the functionality from a shell, everything is behaving as expected, so I assume that everything is installed correctly.
More specifically, when I invoke the following script from a command line:

from opensfm import dataset

datdir='/dataset/Photogrammetry/Lund'

data = dataset.DataSet(datdir)
tracks_manager = data.load_tracks_manager()
TT=tracks_manager.get_all_common_observations("03.jpg","04.jpg")
nr,obs1,obs2=TT[0]

print(obs1.id,obs1.point, obs1.scale, obs1.color)
print(obs2.id,obs2.point, obs2.scale, obs2.color)

I get as output on the command line:
936 [-0.279906 -0.305756] 0.00185086 [146 165 198]
2172 [-0.255517 -0.34403 ] 0.00197027 [166 188 227]

However, when copy-pasting this exact same code into a jupyter cell, I get as output:
936 [-0. -0.] 0.0 [146 165 198]
2172 [-0. -0.] 0.0 [166 188 227]

So, somehow, feature points and scalings are NOT loaded properly, whereas color and id are.

Somehow, I got the impression that it has to be connected to the C-extensions and the environment variables, but I can't be sure. In any case, if I run the above script in a jupyter cell with:
!python3 mytest.py => it gives the right output.
Running in a cell:
%run mytest.py => gives the bad output..

So, if somebody could shed any light on what I am missing, I would be very happy. It's been driving me nuts for some time now..
Cheers.

Reproduce the provided OpenSfM script from a shell, with !python3 mytest.py, and with %run mytest.py in a Jupyter cell. Compare the get_all_common_observations outputs, especially feature points and scales; done means the notebook execution produces the same values as the command-line execution.