Steps verified on Ubuntu 18.04, Cuda 10.2 with a GTX1080 Ti card, and docker 19.03.2 installed on the host machine
The docker image created by Dockerfile is quite large, ~10GB, which means you may need to up the max image size allowed for docker. Check this link for more info. I have tried to streamline the insights from the article and to give the user what they need from it.
Copy daemon.json to your /etc/docker/daemon.json. Stop/start docker:
sudo systemctl stop docker
# sudo rm -rf /var/lib/docker/* ## IF YOU HAVE IMAGES, SAVE THEM IN A TARBALL ARCHIVE AND RESTORE AFTER DOING THIS
sudo systemctl start docker
Note: you may or may not need to do the middle step. If you have previously-built images, you will need to remove them before restarting docker for the updated max image size to take effect.
You can run docker info to verify the desired change took place: Look for the line in the output:
...
Storage Driver: devicemapper
... some lines here ...
Base Device Size: 21.47GB
... more lines follow ...
$ cd {path-to-pvnet}/docker
$ docker build --network=host -t pvnet-nvidia .
This will take awhile...
Follow the steps from the README. Note that the launch-docker.sh script handles mounting the pre-trained model's location on the host inside of the running container instance that the script launches.
$[host] cd {path-to-pvnet}
$[host] ./docker/launch-docker.sh
$[container] source activate pvnet
$[container] cd /home/pvnet/pvnet
$[container] python tools/demo.py
You should see the desired figure pop-up:
And that's it! Once you exit the container, it will automatically be deleted and it's resources released.