akashsengupta1997
commented
3 years ago Hi,
The simplest fix would probably be to use torch==1.4.0 (and torchvision==0.5.0) as stipulated in the README (as well as the appropriate CUDA version - 10.1 perhaps?).
My fork of detectron2 is simply an older version of it. The newer versions of detectron2 are not backwards compatible with older versions of pytorch, hence I recommend using my fork since it works with torch==1.4.0, as does the rest of the code.
If this still doesn't work, you can check out a newer related project here which does not have detectron2 as a dependency (although the code for this project has been tested with torch==1.6.0, so there may be further hassle if the appropriate version of torch is not used).
Akash
AlyanQ
Hi, I've been trying to run your project for the past two days and I can't seem to debug the issue. Your fork of Detectron2 that the project requires never installs. The error is given below.
As you can see, that's not even half of the error output. I have gone back and checked the original Detectron2 repo and tried to debug from there but to no avail since your fork is vastly different from the original. I have applied all the fixes that they mention in the original repo but nothing seems to be working. Writing this issue here as a last resort to see what is wrong.
I am running python=3.7.11, torch=1.8.0, cudatoolkit=11.1 with CUDA version being the same. My nvcc and gcc versions are all compatible with one another as well.