I have just forked Hugin++ from the latest revision of Hugin and fastPTOptimizer from the latest revision of libpano13 and merged my changes into them. The updated sources and the Hugin++ build for Windows can be downloaded here: https://sourceforge.net/projects/huginplusplus/files/development/ The updated fastPTOptimizer is here: https://sourceforge.net/projects/fastptoptimizer/files/
Florian Königstein schrieb am Sonntag, 19. Dezember 2021 um 11:56:16 UTC+1: > I updated the Hugin++ sources and the Windows binaries with David Horman's > latest version of multiblend (and with a bugfix in multiblend concerning a > crash when input images have only transparent pixels): > https://sourceforge.net/projects/huginplusplus/files/development/ > > Florian Königstein schrieb am Mittwoch, 1. Dezember 2021 um 15:56:03 UTC+1: > >> At first, I'd like to say that it's nice that fast geometrical >> optimization is now included in the latest version of libpano >> (libpano13-2.9.21 release candidate rc2). >> The reason why I made my branches of libpano (fastPTOptimizer) and Hugin >> (Hugin++) is the following: >> In the latest version of libpano is one bottleneck that slows down the >> optimization especially for very large panoramas: The calculation of the >> jacobian matrix that is needed by the levenberg marquardt algorithm. >> >> I tested the Hugin 2021.0 beta 1 (Windows binary) and compared it with my >> current version of Hugin++ (in Hugin 2021.0 beta 1 the libpano13-2.9.21 >> release candidate rc1 (not rc2) is used, but it should not make much >> difference in speed). >> >> I performed geometrical optimization for a large panorama with 1351 >> images and 273471 control points (this pano is a bit pre-optimized). >> >> In Hugin 2021.0 beta 1 it took 485 seconds for 5 iterations in strategy 1 >> and 2 iterations in strategy 2. >> >> In Hugin++ the optimization took 63 seconds for 4 iterations in strategy >> 1 and 3 iterations in strategy 2. >> >> So, for this panorama my version is about 8 times as fast as the one in >> libpano for Hugin 2021 beta. >> Probably this speedup factor will get smaller for smaller panoramas. >> >> I know why libpano13-2.9.21 release candidate rc2 is made as it is >> (especially the calculation of the jacobian matrix): It should be better >> maintainable compared to my version of libpano (I use a function >> ChangeAlignParamInImage() in the file adjust.c that does similar jobs than >> SetAlignParams, but ChangeAlignParamInImage() is necessary to avoid the >> bottleneck in the calculation of the jacobian matrix). >> >> I'd like to emphasize that libpano13-2.9.21 release candidate rc2 is >> still a big advantage in speed for large panoramas compared to the version >> libpano13-2.9.20 that is used in Hugin 2020.0.0. In Hugin 2020.0.0 the >> optimization of the above large (pre-optimized) panorama would have taken >> hours. >> >> But in my eyes the speedup factor 8 for my large panorama justifies that >> I keep my fastPTOptimizer parallel to libpano13-2.9.21 and my Hugin++ >> branch parallel to Hugin. >> >> In Hugin++ itself I also made some changes compared to Hugin. In my eyes >> one change (weights for control points) is useful for some "pathological" >> panoramas that are difficult to optimize (e.g. when moving objects like >> leaves of a tree are present and you have no alternative to setting some >> control points onto them). >> In these cases you can tell the geometrical optimizer that some control >> points are less reliable (e.g. those on leaves on a tree): You assign >> weights less than one (1) to them, e.g. 0.1 or 0.01 . On the other hand you >> can assign weights larger than one (e.g. 10 or 100) to control points that >> are more reliable, e.g. those on the trunk of a tree. In fact I used this >> weight feature and could optimize a panorama satisfactorily than I could >> not optimize well without the weight feature. >> >> I'd like to know whether some of you find the "weights for control >> points" feature also useful for some panoramas ? >> >> ChameleonScales schrieb am Samstag, 21. August 2021 um 12:44:19 UTC+2: >> >>> Ok, I was not expecting it to be this complicated. Plus I tried on the >>> new Debian 11 and got an error I didn't see before when configuring hugin++ >>> (CMAKE_DLLTOOL-NOTFOUND) which means the steps may vary depending on >>> the distro and version. >>> Have you considered making an appimage of hugin++? I've never done it so >>> I don't know how easy or complicated it is but if it could be done, any >>> Linux distro could just open huginn++ with nothing more than a double-click. >>> In the current state I prefer to wait for the upcoming features to be >>> implemented in Hugin. Very sorry if if I wasted your time. >>> >> -- 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/95e00a81-b0c7-444a-8d4e-90229ba6dd6cn%40googlegroups.com.
