mmcv-full1.4.0卡wheel解决

实际上就是mmcv和cuda版本不匹配，需要注意先卸载mmdet再下载mmcv-full

look

(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ conda install pytorch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 cudatoolkit=11.3 -c pytorch
(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ pip install -U openmim
(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ mim install mmengine
(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ mim install mmcv-full
(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ pip install -r requirements/build.txt	
(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ python setup.py develop
(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ pip uninstall mmdet
(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ mim install mmcv==1.4.0
(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ pip install -v -e .
(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ pip install mmcv==1.7.2
(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ mim uninstall mmdet
(new_torch) server@server-S8030GM2NE:~/mmdet-rfla$ mim install mmcv-full==1.4.0

一定要先给mmdet干掉在mmcv-full1.4.0