colmap feature extractor error

[hanjoonwon]

i encounterd this error. so i tried ns-process-data images --data images/fox --output-dir output/fox5 --no-gpu this works well but much slower how can i use gpu??

[dberga]

Same error here

F0124 17:45:50.528501 967920 opengl_utils.cc:56] Check failed: context_.create() 
*** Check failure stack trace: ***                                                                   
    @     0x7ff41d56c1c3  google::LogMessage::Fail()                     
    @     0x7ff41d57125b  google::LogMessage::SendToLog()             
    @     0x7ff41d56bebf  google::LogMessage::Flush()                          
    @     0x7ff41d56c6ef  google::LogMessageFatal::~LogMessageFatal()
    @     0x55e513a0430c  colmap::OpenGLContextManager::OpenGLContextManager()
    @     0x55e513909d45  colmap::internal::SiftFeatureExtractorThread::SiftFeatureExtractorThread()
    @     0x55e51390c126  colmap::SiftFeatureExtractor::SiftFeatureExtractor()                                                                                                                             
    @     0x55e513806615  RunFeatureExtractor()
    @     0x55e5137f2eaf  main
    @     0x7ff41b754083  __libc_start_main
    @     0x55e5137f6f6e  _start
Aborted

Running without gpu works though

[dberga]

i encounterd this error. so i tried ns-process-data images --data images/fox --output-dir output/fox5 --no-gpu this works well but much slower how can i use gpu??

I could manage to fix this. What you should do is to compile the colmap on your device. (make sure if you use conda to be on your base environment):

git clone https://github.com/colmap/colmap.git
cd colmap
mkdir build
cd build
cmake .. -GNinja -DCMAKE_CUDA_ARCHITECTURES=native
sudo chown -R $(whoami) .
ninja
sudo ninja install

[hanjoonwon]

i encounterd this error. so i tried ns-process-data images --data images/fox --output-dir output/fox5 --no-gpu this works well but much slower how can i use gpu??

I could manage to fix this. What you should do is to compile the colmap on your device. (make sure if you use conda to be on your base environment):

git clone https://github.com/colmap/colmap.git
cd colmap
mkdir build
cd build
cmake .. -GNinja -DCMAKE_CUDA_ARCHITECTURES=native
sudo chown -R $(whoami) .
ninja
sudo ninja install

Thanks! I have one question addtionally, my desktop colmap version 3.7 and laptop is 3.10 dev when i process custom images, desktop match 88 % but laptop matches 3%... is it normal?

[dberga]

i encounterd this error. so i tried ns-process-data images --data images/fox --output-dir output/fox5 --no-gpu this works well but much slower how can i use gpu??

I could manage to fix this. What you should do is to compile the colmap on your device. (make sure if you use conda to be on your base environment):

git clone https://github.com/colmap/colmap.git
cd colmap
mkdir build
cd build
cmake .. -GNinja -DCMAKE_CUDA_ARCHITECTURES=native
sudo chown -R $(whoami) .
ninja
sudo ninja install

Thanks! I have one question addtionally, my desktop colmap version 3.7 and laptop is 3.10 dev when i process custom images, desktop match 88 % but laptop matches 3%... is it normal?

That might be related to the version or parametization defining how strict is the matching. Low matching (3%) shown in colmap is not wrong. Try same version of colmap to make sure that it is not something related to the computer.

Did you try hloc? Apparently it has a higher matching probability.

[hanjoonwon]

i encounterd this error. so i tried ns-process-data images --data images/fox --output-dir output/fox5 --no-gpu this works well but much slower how can i use gpu??

I could manage to fix this. What you should do is to compile the colmap on your device. (make sure if you use conda to be on your base environment):

git clone https://github.com/colmap/colmap.git
cd colmap
mkdir build
cd build
cmake .. -GNinja -DCMAKE_CUDA_ARCHITECTURES=native
sudo chown -R $(whoami) .
ninja
sudo ninja install

Thanks! I have one question addtionally, my desktop colmap version 3.7 and laptop is 3.10 dev when i process custom images, desktop match 88 % but laptop matches 3%... is it normal?

That might be related to the version or parametization defining how strict is the matching. Low matching (3%) shown in colmap is not wrong. Try same version of colmap to make sure that it is not something related to the computer.

Did you try hloc? Apparently it has a higher matching probability.

Oh is it campatible with nerfstudio? Thanks

[dberga]

i encounterd this error. so i tried ns-process-data images --data images/fox --output-dir output/fox5 --no-gpu this works well but much slower how can i use gpu??

I could manage to fix this. What you should do is to compile the colmap on your device. (make sure if you use conda to be on your base environment):

git clone https://github.com/colmap/colmap.git
cd colmap
mkdir build
cd build
cmake .. -GNinja -DCMAKE_CUDA_ARCHITECTURES=native
sudo chown -R $(whoami) .
ninja
sudo ninja install

Thanks! I have one question addtionally, my desktop colmap version 3.7 and laptop is 3.10 dev when i process custom images, desktop match 88 % but laptop matches 3%... is it normal?

That might be related to the version or parametization defining how strict is the matching. Low matching (3%) shown in colmap is not wrong. Try same version of colmap to make sure that it is not something related to the computer. Did you try hloc? Apparently it has a higher matching probability.

Oh is it campatible with nerfstudio? Thanks

yes, check https://docs.nerf.studio/reference/cli/ns_process_data.html

the --sfm-tool parameter is for selecting either colmap or hloc