I’m started a new thread on the CLIJ Convolution design as the deconvolution one got very long and the discussion about Convolution has different considerations, as now we have to figure out how to make a nice design that can support different types of FFT based filters.
Anyway I’ve added some “proof of concept” code for CLIJ FFT convolution. See the latest commits in ops-experiments.
It is the most “un-DRY” code ever. Next steps are probably to
a) figure out where the long term code should be. Should there be an FFT based repo in the CLIJ ecosystem (as opposed to just cloning parts of ops-exeriments, which is really meant to be for fast prototyping of experimental code)
b) how to organize the code better to avoid repeating so much.
More to come…