Non determinism with tests involving the realign image function
This issue is with regard to the meeting with zesje and WOMM team of 7 june. The issue is that tests involving the function process_page
of scans.py
will sometimes fail and sometimes won't. This happens when using the new realign_image
function introduced in this branch. I've noticed that process_page
will sometimes return the error that it couldn't read the barcode. Upscaling the images from 72 dpi to 144 has resolves this random behavior. It is hard to determine where this problem is exactly since process page uses many complex functions together since realign_image
relies on find_corner_marker_keypoints
and the resulting image is then passed to decode_barcode
so it seems that data can become unreadable if it realigned at lower resolutions.
functions that exhibit this behavior:
test_pipeline
test_rotate [failing_rot]
test_scale [smaller_scale]
test_all_effects [1st full test]
test_all_effects [second full test]
of these failing rot
is interesting because at higher resolutions, process_page
doesn't fail at such high rotations.