Which MMCV version should I use?

[Kev1nZheng]

I try to install mmtracking with cuda11.7 and PyTorch 1.13.1. First, I installed mmcv=2.0.0, and run the demo python demo/demo_mot_vis.py configs/mot/deepsort/sort_faster-rcnn_fpn_4e_mot17-private.py --input demo/demo.mp4 --output mot.mp4 File "/home/xxx/code/mmtracking/mmtrack/__init__.py", line 71 it said f'MMCV=={mmcv.__version__} is used but incompatible. ' \ f'Please install mmcv>=1.3.17, <2.0.0.'

Then I installed mmcv=1.7.1, run the demo File "/home/xxx/miniforge-pypy3/envs/open-mmlab/lib/python3.7/site-packages/mmdet/__init__.py", line 18, in <module> it said f'Please install mmcv>=2.0.0rc4, <2.1.0.'

[tangtang-x]

I'm also having the same issue！

First of all, I configured the environment according to the official document, mmcv==2.0.0, but when I ran the demo, an error was reported： ModuleNotFoundError: No module named 'mmcv.parallel'. After searching, it was found that mmcv2.0 version has dropped 'mmcv.parallel'.

On the second attempt, I recreated the new environment following the MMtracking source code： pytorch==1.10.0 py3.8_cuda11.3_cudnn8.2.0_0 mmcls ==0.25.0 mmcv-full==1.6.1 mmdet==2.3.0 mmengine ==0.8.4 mmtrack ==0.14.0 when I ran the demo, an error was reported：ModuleNotFoundError: No module named 'mmcv.transforms' ，but mmcv.transforms is only available in mmcv2.0.

So how do I configure the environment? Can the author update mmtracking to make it work with mmcv2.0?

[thomasf1]

Got the same issue. The install as per documentation doesn´t work.

[tphlru]

You can install mmcv==2.0.0 and mmengine==latest, just change the control variables in setup.py to 2.0.1 and the same for mmengine and mmdetection. The are my PERSONAL instructions to install mmtrack in conda enviroment(python 3.9):

python 3.9 cuda 121

##for fedora sudo dnf groupinstall 'Development Tools'

wget https://developer.download.nvidia.com/compute/cuda/12.1.0/local_installers/cuda-repo-fedora37-12-1-local-12.1.0_530.30.02-1.x86_64.rpm sudo rpm -i cuda-repo-fedora37-12-1-local-12.1.0_530.30.02-1.x86_64.rpm sudo dnf clean all sudo dnf -y module install nvidia-driver:latest-dkms sudo dnf -y install cuda-12-1

##for fedora

(pythorch 2.1.0) pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu121

echo "export PATH=/usr/local/cuda-12.1/bin${PATH:+:${PATH}}" >> ~/.profile source ~/.profile

pip install 'mmengine' conda install mmcv==2.0.0 pip install 'mmdet==3.0.0rc0'

#change the control versions in setup.py

git clone -b 1.x https://github.com/open-mmlab/mmtracking.git cd mmtracking

#change the control versions in setup.py

pip install -r requirements/build.txt python setup.py develop

[gercorri]

Hi @tphlrussia , we are also getting the same issue. We cannot get mmTracking to work despite trying multiple combinations of versions of different packages and following the above recommendation.

With the above solution, we get a version is not supported for mmengine with mmcv==2.0.0.

@tangtang-x @Kev1nZheng @thomasf1 did either of you manage to get mmTracking working?

We have spent a long number of hours trying various combinations of packages and fixing different errors that came up but ultimately it seems that mmTracking is currently unusable.