MMCV

MMCV is a basic library for computer vision, which supports many open source projects.

warning

RTX 3000 series graphics cards are currently not working properly on PyTorch 1.8.1 (For CUDA 11.1). Please create a PyTorch 1.9.1 version image to use this series of graphics cards. MMCV is not installed in PyTorch 1.10 version.

The official PyTorch image has pre-installed the mmcv-full package, which can be used directly by installing other OpenMMLab toolboxes.

MMDetection

MMDetection is an open source toolbox for target detection based on PyTorch.

MMDetection's version requirements for MMCV see Prerequisites.

# View MMCV version
pip show mmcv-full | grep Version

git clone https://mirror.ghproxy.com/https://github.com/open-mmlab/mmdetection.git
cd mmdetection
# Check out supported versions of MMDetection based on the version of MMCV
git checkout v2.14.0

# Install
pip install -r requirements/build.txt --no-cache
pip install -v -e . --no-cache

Next download a model file for testing.

mkdir checkpoints
cd checkpoints
curl -OL "http://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth"

Open JupyterLab and open the /root/mmdetection/demo/inference_demo.ipynb notepad file in the code you just cloned. Click Run - Run All Cells in the menu.