docker

Guidance to use OpenPCDet with docker

You can either build the docker image through Dockerfile or pull the docker image from dockerhub. Please make sure nvidia-docker is corretly installed and that OpenPCDet is cloned locally.

Build through Dockerfile (WATO)

Build docker image to include local OpenPCDet repository:

docker build -f wato_cu116.Dockerfile -t openpcdet-docker /home/{USER}/

Then login to the container and make sure to mount the path of the model and data as volumes:

docker run -it -v {WATO_DRIVE_PATH_TO_DATA}:/data/ -v {WATO_DRIVE_PATH_TO_MODELS}:/models openpcdet-docker:latest /bin/bash

E.g.

docker run -it -u root -v /mnt/wato-drive2/nuscenes-1.0-mini:/data/nuscenes-mini -v /mnt/wato-drive2/perception-weights/lidar_object_detection:/models openpcdet-docker:latest /bin/bash

Inside the container, you can run OpenPCDet through xvfb using the start-xvfb.sh script:
where the models and data directories are from the root of your container and must be downloaded from their respective sources!

cd tools
/usr/local/bin/start-xvfb.sh python3 demo.py --cfg_file cfgs/kitti_models/pv_rcnn.yaml --ckpt /models/pv_rcnn_8369.pth --data_path /data/velodyne/data/0000000050.bin

The rendered image will be created in the same directory as demo.py, this can be copied from the docker container where the second argument can be customized to your desired path:

docker cp {CONTAINER_ID}:/OpenPCDet/tools/bb_output.png /home/{USER}

Pushing to Github GHCR

To manually push an image to GHCR, first login to GHCR:

docker login ghcr.io

Then tag and push the image:

docker tag openpcdet-docker:yourtag ghcr.io/watonomous/openpcdet-docker:yourag
docker push ghcr.io/watonomous/openpcdet-docker:yourtag

Build Through Dockerfile

Build docker image that support OpenPCDet through:

docker build ./ -t openpcdet-docker

Note that if you would like to use dynamic voxelization, you need further install torch_scatter package.

From this Dockerfile, the installed version of spconv is 2.x, if you would like to use spconv 1.2.1, please follow these steps:

git clone -b v1.2.1 https://github.com/djiajunustc/spconv spconv --recursive
cd spconv
python setup.py bdist_wheel
cd ./dist
pip install *.whl

Pull From Dockerhub

Run the following script to pull the docker image:

docker pull djiajun1206/pcdet:pytorch1.6