@haesleinhuepf Hi mate. I'n playing with the RL deconv on GPU in CLIJ.
I noticed that the imageJ pixel data types need to be the same (32 bit works) for the source image and the kernel image (PSF), for the result image to be ok. If they are different I get a garbage result image.
Is that enough info, can you reproduce?
These fail: 8 bit vs 16 bit, 32 bit vs 16 bi, 16 vs 16
Only 32 bit vs 32 bit seems to work so far for me.
B.T.W.
my GPU selection is
Quadro M1000M
and
Intel(R) HD Graphics 530
Both work, but you can guess which one is faster ;-)
using openCL device as CPU: intel i7-6820HQ also works fine, as you would expect.
@haesleinhuepf Hi mate. I'n playing with the RL deconv on GPU in CLIJ. I noticed that the imageJ pixel data types need to be the same (32 bit works) for the source image and the kernel image (PSF), for the result image to be ok. If they are different I get a garbage result image. Is that enough info, can you reproduce? These fail: 8 bit vs 16 bit, 32 bit vs 16 bi, 16 vs 16 Only 32 bit vs 32 bit seems to work so far for me.
B.T.W. my GPU selection is Quadro M1000M and Intel(R) HD Graphics 530 Both work, but you can guess which one is faster ;-) using openCL device as CPU: intel i7-6820HQ also works fine, as you would expect.