You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Describe the bug
The two Gaussian convolution layers, images.GaussConvFourier and images.GaussConvImage, do not yield consistent results when considering elliptical beams with some non-zero rotation value Omega.
It appears the images.GaussConvFourier routine is consistent with our convention expectations (positive Omega is rotated from North towards East, 0 degrees has FWHM_maj aligned with N-S) whileimages.GaussConvImage follows the opposite convention.
See the output of the images.test_GaussConvFourier_rotate and images.test_GaussConvImage_rotate tests.
Screenshots
The routines appear to be consistent for Omega = 0.
But look noticeably different for Omega = 30. Compare the struts/veins in the lower half of the butterflies wings in the right vs. left wings. The GaussConvImage appears to give the expected result, where struts aligned with the beam major axis look sharpest.
Suggested fix
Investigate what is flipped inside the GaussConvFourier routine with regards to the beam.
Write a new test directly comparing the per-pixel output from images sent through both routines, for some non-zero and non-symmetric rotation.
Also, the Omega=0 results look less consistent than expected, so further tests to investigate these discrepancies.
The text was updated successfully, but these errors were encountered:
Describe the bug
The two Gaussian convolution layers,
images.GaussConvFourier
andimages.GaussConvImage
, do not yield consistent results when considering elliptical beams with some non-zero rotation value Omega.It appears the
images.GaussConvFourier
routine is consistent with our convention expectations (positive Omega is rotated from North towards East, 0 degrees has FWHM_maj aligned with N-S) whileimages.GaussConvImage
follows the opposite convention.See the output of the
images.test_GaussConvFourier_rotate
andimages.test_GaussConvImage_rotate
tests.Screenshots
The routines appear to be consistent for Omega = 0.
But look noticeably different for Omega = 30. Compare the struts/veins in the lower half of the butterflies wings in the right vs. left wings. The GaussConvImage appears to give the expected result, where struts aligned with the beam major axis look sharpest.
Suggested fix
GaussConvFourier
routine with regards to the beam.The text was updated successfully, but these errors were encountered: