scatter estimation problem in recon_test_pack

[KrisThielemans]

Originally observed in https://github.com/NikEfth/STIR/pull/66#issuecomment-1878475159 but this is not a TOF issue, and happens on current master.

Changing the template to DSTE span 2, max rd3 (from span 0, max rd2) encovers a problem in the scatter estimation, i.e. run_scatter_tests.sh fails with too large difference between simulation and estimation. This is pretty weird as segment 0 for both these templates should be identical.

Initial investigation shows that

• scatter simulation and upsampling is the same
• reconstructed images during scatter estimation look fine but have different scale (and both seem to be wrong)

estimated (left) and simulated (right) scatter and diff (bottom) (segment 0, central view): reconstructed images during scatter (from span=2 left, from span=0 right, input = bottom) ROI values of these images are ~12, ~8, 1 resp. I'd expect a factor 4 due to the ratio in x-voxel size between reconstructions and input though.

This is all weird, as the reconstructed images should have the same values, and it's hard to understand why the difference in seg 0 is on one end of the scanner, while actually the axial image profiles are fine.

[markus-jehl]

This looks vaguely related to axial geometry fixes in the downsampling done here: https://github.com/UCL/STIR/pull/1466

So maybe this issue can be closed?

[KrisThielemans]

I have no idea. We should re-check this, and I have no time for it.