hi, 1) 16s in single-thread python sounds like it would be possible to do < 100ms in a programming language.
4) that depends on the order of modules. if you want to do it in the current sharpen module as an option, it'll come pretty much last. if you do it early, dt will transparently cache the output for you and do the other computations in darkroom mode on top. that said, i doubt you want to implement this for raw/bayer/xtrans images or run it before denoising. 5) indeed this imagej plugin has interesting code, albeit java (but GPLv3, too). the results look great and it even has a spatially varying psf version. the test cases on this web page seem to be more in the 256^2 range of problem size, so the speed issue will need investigation. i suppose opencl may help here. cheers, jo On Thu, Oct 12, 2017 at 5:23 PM, Aurélien PIERRE <rese...@aurelienpierre.com> wrote: > Hi ! > > I understand the performance concerns and I'm working on some trade off. > But… > > 1 - Non-blind Richardson-Lucy deconvolution by gradient descent with Total > Variation regularization (probably not the algorithm used in Gimp since it's > relatively recent) gives very good results in 25 iterations, thus 16 s on my > 2 Mpx test picture with a Python script (on an i7 Ivy bridge laptop). > > 2 - Myopic deconvolution (the sort of "blind" deconvolution where you give a > good-enough initial guess of the blur profile) converges faster than the > blind one, > > 3 - The most computation-demanding operation is convolution product (2 > FFT-convolve by non blind iteration, 4 by blind iteration). The good news is > we don't need to compute them on the whole picture (it's actually bad when > you have a large bokeh area) and you can/should mask the area of interest > and do the computations only on it. It saves a lot of time and gives better > results on some cases. > > 4 - It should be possible to deconvolve the RAW pic first, cache it, then > apply the further edits on the cached picture (similar to the HDR workflow). > > 5 - Piccur uses a myopic deconvolution (from what I have understood), and > seems to offer rather decent time/quality ratio. Also, ImageJ has a similar > open-source plugin (http://imagej.net/Parallel_Iterative_Deconvolution) > which code could be of interest. > > Thanks for your interest ! > > Aurélien PIERRE > aurelienpierre.com > > ________________________________ > > Le 2017-10-11 à 14:59, Heiko Bauke a écrit : > > Hi, > > Am 11.10.2017 um 19:11 schrieb Martin Marmsoler: > > Gimp use python as scripting language. It might be easier to port for Gimp? > > > by the way: there is a Richardson Lucy sharpening filter in G'MIC. (As far > as I understand this is a non-blind deconvolution algorithm.) > > > Heiko > > > > ___________________________________________________________________________ > darktable developer mailing list to unsubscribe send a mail to > darktable-dev+unsubscr...@lists.darktable.org ___________________________________________________________________________ darktable developer mailing list to unsubscribe send a mail to darktable-dev+unsubscr...@lists.darktable.org
