Closed jytime closed 5 months ago
Okay after using the cmake command of
cmake .. -DCMAKE_CUDA_ARCHITECTURES=80 -DCMAKE_INSTALL_PREFIX=/data/home/jianyuan/ptcolmap/dependencey/colmap/build/colmap_installed/ -DCMAKE_BUILD_TYPE=Debug -DGUI_ENABLED=OFF
and only running make -j && make install
for the following modifications, the re-compiling time of colmap is reduced to under a minute. Then we need to rerun "python -m pip install -e ." for pycolmap (every time after re-compiling the C++ code of colmap), which will take around 2 minutes. Although it is still a bit long, I share the command here in case someone may also need it.
Using ninja can decrease the build time further: cmake .. -GNinja [...] && ninja install
Hi all,
Thanks for your great work. I am trying to build a project on top of pycolmap (and hloc). I have successfully compiled colmap 3.9.1 and pycolmap 0.5.0 from source.
As I delve deeper, I am trying to find efficient ways to debug the code. For instance, while trying to understand the process within pycolmap.verify_matches(), I traced back to EstimateTwoViewGeometry in COLMAP. To check the H_F_inlier_ratio for various runs, I added the following logging codes:
However, this approach requires recompiling COLMAP from source, which takes approximately 30 minutes for each modification. I am wondering if there is a more efficient way to debug the code without having to recompile COLMAP each time. As I am relatively new to building projects with C++, I apologize if this question seems elementary. Any guidance or suggestions would be greatly appreciated.
Best Regards, Jianyuan