search

facebookresearch / consistent_depth

We estimate dense, flicker-free, geometrically consistent depth from monocular video, for example hand-held cell phone video.
MIT License
1.62k stars 235 forks source link

pytorch/cuda version #8

Closed mattclifford1 closed 3 years ago

mattclifford1 commented 4 years ago

I've noticed that pytorch version 1.4 with cuda 10.0 is specified in the requirements (torch==1.4.0+cu100). Is there any specific reason for this requirement or can I run the code with more up to date pytorch and/or cuda?

Thanks, Matt

miaoqiz commented 4 years ago

I ran into this issue when building "colmap" -

" nvcc fatal : Unsupported gpu architecture 'compute_75' "

After using a docker container that has cuda-10.0, "colmap" is built correctly.

Thanks!

mattclifford1 commented 3 years ago

I've actually found that colmap builds fine with cuda 10.1. However it's important to build with gcc-8 or lower as the build will fail with gcc-9.

This means changing the code in ./scripts/install_colmap_ubuntu.sh on line 47 from cmake .. to CC=/usr/bin/gcc-8 CXX=/usr/bin/g++-8 cmake ..

or substituting version 8 for a lower version depending on your setup.