| Git Clone URL: | https://aur.archlinux.org/colmap-git.git (read-only, click to copy) |
|---|---|
| Package Base: | colmap-git |
| Description: | General-purpose Structure-from-Motion (SfM) and Multi-View Stereo (MVS) pipeline with a graphical and command-line interface. |
| Upstream URL: | https://colmap.github.io/ |
| Keywords: | 3d multi-view-stereo reconstruction structure-from-motion |
| Licenses: | GPL |
| Conflicts: | colmap |
| Provides: | colmap |
| Submitter: | bartus |
| Maintainer: | bartus (mgrewe) |
| Last Packager: | bartus |
| Votes: | 5 |
| Popularity: | 0.000000 |
| First Submitted: | 2017-03-21 10:48 (UTC) |
| Last Updated: | 2026-04-15 14:20 (UTC) |
Something seems to be wrong in the makefile: nvcc fatal : Unsupported gpu architecture 'compute_20' CMake Error at pba_generated_ProgramCU.cu.o.Release.cmake:219 (message): Error generating /tmp/yaourt-tmp-herbert/aur-colmap-git/src/build/lib/PBA/CMakeFiles/pba.dir//./pba_generated_ProgramCU.cu.o
full log: http://ix.io/1Bi8
Thanks @bartus. The build that you linked me to worked like a charm!
@verbal007: use prev version of PKGBUILD. Arch have unbundled dependencies of freeimage, where manjaro steel ships all deps inside freeimage package.
Let see if manjaro follows arch, otherwis will make package detect os and adjust accordingly.
A few weeks back I was able to run this build with no problems on Manjaro. Today I'm unable to complete the build. The error I'm getting is:
==> Building and installing package
==> Install or build missing dependencies for colmap-git:
error: target not found: jxrlib
@NicolasV: I made dependencies less restrictive, now build works with other CUDA versions. Noticed that build complains of CUDA loss than 7.0 -- Disabling CUDA support (found version 6.5 but >= 7.0 required) so building with cuda65 perhaps won't work.
You will have to patch CUDA_MIN_VERSION in CMakeList.txt
104 set(CUDA_MIN_VERSION "7.0")
105 if(CUDA_ENABLED)
106 find_package(CUDA ${CUDA_MIN_VERSION} QUIET)
107 endif()
I have a comment. I don't have a recent GPU (GTX 460) so I use "cuda65" instead of "cuda" So I have to edit the PKGBUILD in order to enable CUDA during the compilation. PKGBUILD should detect either "cuda" or "cuda65". Thank you
@Djangata: cuda-9 had downgraded cuda-host-compiler to gcc5 from previous gcc6, which causes the issue. Should be fixed now.
i get an error and don't really know how to troubleshoot it
[ 13%] Linking CXX static library libestimators.a [ 15%] Built target estimators [ 17%] Built target util_cuda [ 17%] Built target flann [ 39%] Built target graclus [ 41%] Built target lsd [ 41%] Building NVCC (Device) object src/ext/PBA/CMakeFiles/pba.dir/pba_generated_ProgramCU.cu.o gcc: error: unrecognized command line option ‘-fno-plt’ CMake Error at pba_generated_ProgramCU.cu.o.Release.cmake:219 (message): Error generating /home/jordan/Programs/Colmap/colmap-git/src/build/src/ext/PBA/CMakeFiles/pba.dir//./pba_generated_ProgramCU.cu.o
make[2]: [src/ext/PBA/CMakeFiles/pba.dir/build.make:65: src/ext/PBA/CMakeFiles/pba.dir/pba_generated_ProgramCU.cu.o] Error 1 make[1]: [CMakeFiles/Makefile2:647: src/ext/PBA/CMakeFiles/pba.dir/all] Error 2 make: *** [Makefile:130: all] Error 2 ==> ERROR: A failure occurred in build(). Aborting...
Please build from #branch=master rather than from #branch=dev by default. Master seems to be more stable for everyday use.
@Harvie: Done, mind that dense reconstruction is only available on CUDA (no CPU support planned). You will have to use one of those #{mve,pmvs,openmvg,openmvs}.
Pinned Comments
bartus commented on 2019-04-10 11:42 (UTC) (edited on 2026-04-29 21:18 (UTC) by bartus)
This package is also hosted on GitHub.
issuesandpatchesat GitHub.Use env vars to control build process:
BUILD_CUDA=ON|OFFto skip cuda kernel build (default 'ON')CUDA_ARCH="5.2,7.2"to build for a specific Cuda arch, supports multiple values.CMAKE_FLAGS="xxx:yyy"to add extra CMake switches (colon-separated list)Usage cases:
export BUILD_CUDA=OFFbefore buildBUILD_CUDA=OFF ~your-aur-helper~makepkg BUILD_CUDA=OFFyay -S colmap-git --mflags "BUILD_CUDA=OFF"