Open TNodeCode opened 8 months ago
Because it's hard to find, I thought I'd mention the code for the 1.6 version of MMCV-full is here: https://github.com/open-mmlab/mmcv/releases?page=2 https://github.com/open-mmlab/mmcv/tree/v1.6.0
Once this was built, the installation of the spine detection code was effortless.
The current installation script provided in this repository doesn't work anymore with PyTorch 2.x and CUDA 12.x. The problem is that the mmcv-full library in the PyPi repository wasn't compiled for CUDA 12 / PyTorch 2.x. You can fix this by uncommenting the line
mim install opencv-full
in the fileinstall_requirements.sh
and downloading the source code of mmcv-full from this repository: https://github.com/open-mmlab/mmcv Switch to the1.x
branch, otherwise you would install mmcv 2.x. Then run the commandFORCE_CUDA=1 MMCV_WITH_OPS=1 pip install -e . -v
from the mmcv directory. If you want to install this library on a cluster using Slurm make sure to create a bash script for the installation process that can then be executed viasbatch
.