RDP gradient can generate NaNs if epsilon is zero
KrisThielemans opened this issue · 2 comments
KrisThielemans commented
Reconstructing the mMR NEMA data gives me NaNs in the background of the image, then quickly ruining everything of course. Obviously, when epsilon
=0, the function is not differentiable at x,y=0. We try to set the gradient to zero, but apparently still miss some cases. Relevant code is
STIR/src/recon_buildblock/RelativeDifferencePrior.cxx
Lines 403 to 424 in 8ced2d7
@Imraj-Singh have you seen this?
robbietuk commented
Could be because of small x/y values?
KrisThielemans commented
This was fixed by #1410