Fix for bug in ray tracing projection which produced axial artifacts due to flipped theta angle

[markus-jehl]

Fixes #1223

[KrisThielemans]

Do we now test with segment !=0 ? You'd have hoped that this would have shown the original mistake.

[markus-jehl]

Do we now test with segment !=0 ? You'd have hoped that this would have shown the original mistake.

Ah, I forgot to mention this! I tried to modify @danieldeidda 's test to include non-zero segments and look at non-trivial entries in the proj data. Unfortunately, even though the proj data look very similar, the bin values still vary in too many places to compare the proj data bin by bin (which might be why the test currently only looks at segment 0, axial_pos 0, tangential_pos 0 and then loops through the views). So I'm still unsure how to best test this without simplifying too much.

[KrisThielemans]

I tried to modify @danieldeidda 's test to include non-zero segments and look at non-trivial entries in the proj data. Unfortunately, even though the proj data look very similar, the bin values still vary in too many places to compare the proj data bin by bin (which might be why the test currently only looks at segment 0, axial_pos 0, tangential_pos 0 and then loops through the views).

sorry, could you elaborate a little bit? Which test was this?

Otherwise, this is good to go, as soon as you add something to the release notes of course!

[markus-jehl]

I tried to modify @danieldeidda 's test to include non-zero segments and look at non-trivial entries in the proj data. Unfortunately, even though the proj data look very similar, the bin values still vary in too many places to compare the proj data bin by bin (which might be why the test currently only looks at segment 0, axial_pos 0, tangential_pos 0 and then loops through the views).

sorry, could you elaborate a little bit? Which test was this?

Otherwise, this is good to go, as soon as you add something to the release notes of course!

I mean the run_map_orientation_test in test_blocks_on_cylindrical_projectors.cxx. I have locally modified it to call set_blocks_projdata_info instead of set_direct_projdata_info, and then write out the projdata1 and projdata2 to compare them visually. The obvious flipped vertical lines are gone, but if I extend the bin comparison to use non-trivial positions there are some bins with large differences.

[KrisThielemans]

if I extend the bin comparison to use non-trivial positions there are some bins with large differences

hmmm. that isn't reassuring!

Anyway, let's merge this then and create an issue for that. squash-merge ok?

[markus-jehl]

if I extend the bin comparison to use non-trivial positions there are some bins with large differences

hmmm. that isn't reassuring!

Anyway, let's merge this then and create an issue for that. squash-merge ok?

Sure! I'll create the issue right away