Before committing to 1.1.0, I have decided to release another lux binary in the 1.0.9 series, because I have put quite some work into lux since 1.0.9a and I'd like to have the software in use for some time to see if any bugs crop up. My 'grooming session' is now done, and I have documented all options in lux_options.html. Find new binaries on my bitbucket site's download page:
portable windows version: https://bitbucket.org/kfj/pv/downloads/lux_for_windows_1.0.9b.zip windows installer: https://bitbucket.org/kfj/pv/downloads/lux_1.0.9b_setup.exe debian package: https://bitbucket.org/kfj/pv/downloads/lux-1.0.9b-amd64.deb So, what's new? Casual users likely won't notice any changes - the focus is more on ironing out bugs and glitches. lux is complex software, and sometimes changes in one part affect other parts in unwanted ways - for example, the comic book mode went dysfunctional, because I changed the semantics of the initial_dx and initial_dy options (that's fixed now). The truly new features are quite 'specialist': it's about creating image pyramids, which are central to both the viewer and the lux version of the Burt and Adelson image splining algorithm. What I added here are new low-pass filters which are used to create a pyramid level from the level 'below' it. Until now, lux was using b-spline reconstruction kernels, 'area' decimation and a small binomial (.25, .5, .25). Now I added the next odd binomial, an 'optimal' Burt filter, and FIR half-band filters based on a truncated sinc with a Hamming window, as explained here: https://www.dsprelated.com/showarticle/1113.php The filters are used in a 'convolving basis functor', a hybrid between b-spline evaluation and FIR filter which allows off-grid access. They require a spline pyramid to operate on, which makes the pyramid building code a bit slower. Especially these filters should provide near-optimal results - for more on the topic, please visit the documentation - or discuss it here! Visibly, the change is hard to detect, because lux' image pyramids were pretty good as they were. If you want to give it a try, invoke lux with --pyramid_smoothing_level=-7 for a 7-tap half-band or -11 for an 11-tap half-band etc. (the half-band filters lux uses have tap counts of 4N-1). If you (optionally) combine that with --decimate_area=yes and zoom out (start with 1:1 magnification by pressing '1', then zoom out) the resulting animation should be practically free from pyramid level switching artifacts. One more bit of good news: I finally managed to find someone who has a recent intel core-i processor which offers the AVX512 instruction set, and managed to talk him into running lux on it. And, lo and behold, the AVX512 capability was detected and the AVX512-specific code used! So far I have no idea if there is a significant performance gain from using the new ISA. but I suspect that the doubled register size vs. AVX2 should make a fair difference, because a lot of lux code is running in the registers rather than accessing RAM. Any of you out there with AVX-512 capable processors, please check it out! Kay On Monday, July 5, 2021 at 10:35:35 AM UTC+2 kfj wrote: > Dear all! > > I've made an intermediate release of lux with a few bug fixes and changes > to the docu which have accrued during the past few weeks, during my > 'grooming' session, which will continue until the release of lux 1.1.0. The > intermediate release has version lux_1.0.9a, downloadable binaries can be > found in the Downloads section of the bitbucket repo: > > https://bitbucket.org/kfj/pv/downloads/lux-1.0.9a-amd64.deb > https://bitbucket.org/kfj/pv/downloads/lux_1.0.9a_setup.exe > https://bitbucket.org/kfj/pv/downloads/lux_for_windows_1.0.9a.zip > > And the README in HTML format was also updated: > > https://bitbucket.org/kfj/pv/downloads/README.html > > The windows installer has seen some work (thanks, Bernd!): > > - it's less obtrusive, per default, lux is not launched after installation > - for several extensions, lux can now be chosen as target via the context > menu > - SendTo can be used to invoke lux with a selection from the explorer > > Kay > On Wednesday, June 23, 2021 at 11:14:22 AM UTC+2 kfj wrote: > >> Dear all! >> >> I'm happy with the level of functionality in lux 1.0.9, and since I've >> seen no complaints - either here or as issues on the bitbucket page - I >> can only assume that lux users agree ;) >> >> I'm now taking the time to consolidate: I'm going through the various >> options (as represented by command line arguments), looking at each in >> turn, trying out whether they (still) work as intended, debugging as I >> go along, and producing more documentation. I have started on a new HTML >> document, which gives a lexical listing of lux options - rather than the >> more thematic grouping in the README file - and this is progressing >> nicely and has already reached the letter 'F'. You can find the source >> code of this document as restructured text in the repo, and an HTML >> rendition on my bitbucket.io page: >> >> https://kfj.bitbucket.io/lux_options.html >> >> I'll update the HTML version as lux_options.rst grows. I thought it a >> good idea to announce this text early, and I'd like to invite you to >> have a look at the text, and to comment on it - especially if I don't >> make sense or use my own jargon - as a developer, I may have come to use >> terms which make sense for me, but aren't comprehensible without >> explanation. >> >> Kay >> > -- A list of frequently asked questions is available at: http://wiki.panotools.org/Hugin_FAQ --- You received this message because you are subscribed to the Google Groups "hugin and other free panoramic software" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/hugin-ptx/ac3939ed-6a3f-43e9-80d8-c4acd89db1f1n%40googlegroups.com.
