Small heads-up. After more checking, and thanks to PanoStitch source, I was able to find out why cpfind/autooptimiser/... were randomly freezing. Turns out I still don't know the exact cause, but it seems linked to the way Java Runtime is calling binaries. If I ran cpfind manually, it would go all the way through perfectly fine. If I ran it through the Runtime class of Java, it would freeze. I ended up writing a small helper class in JNI that calls C's popen function to call every binary, and it seems to work fine now (at least, no more random freeze).
My rendering chain is now like this: 1. pto_gen -p2 2. cpfind --multirow 3. ptclean 4. autooptimiser -a -l -m -s 5. pano_modify 6. nona 7. multiblend I swapped enblend for multiblend, which is noticeably faster than enblend for the same quality. After checking cpfind parameters and docs on PanoTools Wiki, it looks like there is a "--prealigned" mode, which matches images after a predefined order (and thus, much faster and in a more exact way). Since I'm using the gyroscope to give out a preview of the final sphere, there should be a way to store each image supposed position at the time of capture, and feed it to cpfind. Unless I understand it wrong, this should make cpfind much more precise and faster, right? Unfortunately, I couldn't find any documentation or example showing how/what to write in the pto to pre-align images (or what pto_xxxx tool could be used for that). Could someone give me some hint about that? Le jeudi 1 août 2013 17:54:02 UTC+2, Matthew Petroff a écrit : > > They replied, providing the source code at this link: > https://www.dropbox.com/s/adqbtaej8gud6wn/PanoStitch-source.7z > > -Matthew > > On Thursday, August 1, 2013 9:49:19 AM UTC-4, Guillaume Lesniak wrote: >> >> I've sent one too yesterday actually. He said he's currently on vacations >> and he will send it next week. >> >> Anyway, back on topic :) Is there any known issue that would cause >> ptclean or autooptimiser to hang under certain conditions? >> >> Guillaume >> >> Le jeudi 1 août 2013 15:44:10 UTC+2, Matthew Petroff a écrit : >>> >>> I've sent an email to the developer requesting the PanoStitch source >>> code. We'll see if he complies. >>> >>> -Matthew >>> >>> On Wednesday, July 31, 2013 11:58:41 AM UTC-4, Guillaume Lesniak wrote: >>>> >>>> Confirmed: it uses Hugin GPL code in libpano.so: >>>> >>>>> # strings libpano.so | grep "jni" >>>>> >>>>> jni/prebuilts/include/boost/exception/detail/exception_ptr.hpp >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/panodata/Panorama.h >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/hugin_utils/stl_utils.h >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/nona/Stitcher.h >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/nona/Stitcher.cpp >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/panodata/Panorama.cpp >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/hugin_utils/utils.h >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/panodata/image_variables.h >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/panodata/PanoramaOptions.cpp >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/panodata/PTScriptParsing.cpp >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/panodata/SrcPanoImage.cpp >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/panodata/ImageVariableGroup.cpp >>>>> >>>>> >>>>>> jni/hugin-2011.4.0/src/hugin_base/panodata/StandardImageVariableGroups.cpp >>>>> >>>>> jni/hugin-2011.4.0/src/hugin_base/panotools/PanoToolsInterface.cpp >>>>> >>>>> >>>>> >>>> >>>> Le mercredi 31 juillet 2013 17:52:11 UTC+2, Guillaume Lesniak a écrit : >>>>> >>>>> There is no official way to try it without CM, but I might try to >>>>> build it externally in some time. I'll post it here if I have time. >>>>> >>>>> If PanoStitch uses Hugin, then it's an illegal app. The SIFT >>>>> algorithms are patented by the University of British Columbia, and >>>>> everything is GPL, which means the app must be open-sourced too, >>>>> especially >>>>> the changes made to it (and Hugin is far from compiling on Android >>>>> without >>>>> changes). PanoStitch is not only a paid app, but also not releasing AT >>>>> LEAST their changes, making it a 100% illegal app. You might want to >>>>> bring >>>>> that to the GPL-Violations list. >>>>> That's why I put Focal as GPL too, and that's why I was excited to >>>>> work on it: because I could use basically everything I want without >>>>> worrying about licensing. >>>>> >>>>> I'll eventually check and compare the speed, but if they made no >>>>> changes, it'll likely be the same as I have. I'll unpack/decompile the >>>>> APK >>>>> for further licensing concerns. >>>>> >>>>> >>>>> >>>>> Le mercredi 31 juillet 2013 16:34:03 UTC+2, Jim Watters a écrit : >>>>>> >>>>>> Guillaume, >>>>>> >>>>>> Great news! >>>>>> I've watched a video demo of Focal but have not tried it out yet. >>>>>> >>>>>> PanoStitch, available on Google Play ( there is a free demo and a >>>>>> paid full version) >>>>>> >>>>>> https://play.google.com/store/apps/details?id=com.myboyfriendisageek.panostitch.demo&hl=en >>>>>> Also uses Panotools and Hugin but I don't notice any credit or link >>>>>> to the source code they are using. >>>>>> It will give you a chance to compare speeds and see if they managed >>>>>> to improve it. >>>>>> >>>>>> Developers, >>>>>> >>>>>> What are the GLP requirements for using Panotools and Hugin in an >>>>>> Android app? >>>>>> >>>>>> Jim Watters >>>>>> >>>>>> On 2013-07-31 7:06 AM, Guillaume Lesniak wrote: >>>>>> >>>>>> Hey everyone, >>>>>> >>>>>> As you might or might not know, I've released a component called >>>>>> "PicSphere" in the new CyanogenMod camera app Focal. This component is >>>>>> based on PanoTools and Hugin libraries. >>>>>> I've been mostly hacking around to make it work, and it does work, >>>>>> but there are still a few kinks I need to iron out. >>>>>> >>>>>> The repositories are available at >>>>>> https://github.com/cyanogenmod/android_external_Focal >>>>>> >>>>>> - First, cpfind is just crashing and I cannot get it to work. I >>>>>> don't have the stacktrace anymore, as I ended up using autopano-sift-c >>>>>> for >>>>>> the alignment and cp detection, but I read here and there that cpfind >>>>>> works >>>>>> better (as in, faster) than autopano is most situations. >>>>>> - I had to hack around a bit that wouldn't build in vigra, see >>>>>> https://github.com/CyanogenMod/android_external_Focal/blob/cm-10.2/vigra/include/vigra/impexbase.hxx#L154 >>>>>> - >>>>>> it doens't seem to affect the final result, but I would like to know if >>>>>> there's a better way. Basically, the compiler cannot find >>>>>> Accessor::ElementAccessor >>>>>> - Another big problem is that it's slow. Stitching a correct panorama >>>>>> with 640x480 shots takes about a minute, where it takes 3 seconds on my >>>>>> home PC. I know smartphones aren't as powerful as PCs, but going from 3 >>>>>> seconds to 1 minutes is quite a leap (for a quadcore 1.5GHz, 2GB RAM >>>>>> device). Is there any plans for NEON/ARM enhancements, or at least could >>>>>> you give me some pointers on where should I focus optimizing (adding >>>>>> NEON >>>>>> ASM code) to make the whole process faster. >>>>>> >>>>>> Here's the flow of the rendering process (from >>>>>> https://github.com/CyanogenMod/android_packages_apps_Focal/blob/cm-10.2/src/org/cyanogenmod/focal/picsphere/PicSphere.java#L183) >>>>>> >>>>>> : >>>>>> >>>>>> >>>>>> 1. autopano --align --bottom-is-left --generate-horizon 2 >>>>>> --ransac on --maxmatches 30 --keep-unrefinable off --projection >>>>>> 2,HFOV >>>>>> 2. ptclean *--- I disabled it, as it seems to hang for panoramas >>>>>> above 180°. I tracked the bug down to some function pointer call that >>>>>> never >>>>>> returns, but didn't go any further.* >>>>>> 3. autooptimiser -v HFOV -p >>>>>> 4. pano_modify -o project.pto --center --canvas=3000x1500 >>>>>> 5. nona -o project -m TIFF_m >>>>>> 6. enblend *.tif >>>>>> >>>>>> Note: On step 4, I use a fixed 3000x1500 canvas for the simple reason >>>>>> that Google PhotoSphere XMP metadata requires 2:1 ratio pictures. I plan >>>>>> to >>>>>> hack the sources further and enforce a 2:1 ratio after calculating the >>>>>> crop >>>>>> size, to keep the real image size. >>>>>> autooptimiser alignment also seem to fail if there is any >>>>>> unconnected image. >>>>>> >>>>>> What do you think of this flow? Could it be better? >>>>>> >>>>>> My primary concern remains the speed of the rendering, so if anyone >>>>>> would like to help me or take the task, I'd be more than happy. >>>>>> >>>>>> Thanks for your help! >>>>>> >>>>>> >>>>>> -- >>>>>> Jim Wattershttp://photocreations.ca >>>>>> >>>>>> -- 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/04875880-d19d-4863-b522-bcdccb2ad2c5%40googlegroups.com. For more options, visit https://groups.google.com/groups/opt_out.
