Hi,
I'm glad that the syntax is now right - we agree that the
help is unclear and have fixed it for the next release/patch.
As for timing, it will use either fft or direct image-space
convolution depending on an empirical test that we
have inserted that selects the fastest option based on our
benchmarks. With such a large kernel it is very likely
that it will use fft.
The warning is to do with the fact that for odd-sized
kernels an arbitrary decision is made with respect to
where the "centre" should be, and so it just alerts the
user to the fact that the convolved image might be
slightly offset from where they are expecting it to be
(given that we make the output image the same
size as the input image).
Hope this makes sense.
All the best,
Mark
Joerg Magerkurth wrote:
>Hi,
>
>I thought in "-kernel file" the word "file" has to replaced by the real
>filename, it is not clear in the documentation, that file is a keyword for
>avwmaths++. With your last answer I find my fault: "-kernel file MYFILE",
>now it works.
>
>But a warning occurs: Off_centre convolution being performed as kernel has
>even dimendions
>
>What does it mean? My kernelfile is isometric hamming 32x32x32 normalized.
>
>Does use avwmaths++ use fft to perform the computation, it takes a long time.
>
>Joerg
>
>
|