How to deploy UFlow

Get the forked repo

git clone [email protected]:jcaballeros/google-research.git

Get the dataset

cd google-research/uflow
wget http://files.is.tue.mpg.de/sintel/MPI-Sintel-complete.zip

Create Docker image

docker build -t jcaballero/uflow .

Create container

docker run --gpus all --name uflow-container -it jcaballero/uflow bash

or to use only one GPU, say GPU 1:

docker run --gpus '"device=1"' --name uflow-container -it jcaballero/uflow bash

Restart container in exit status

docker start uflow-container

Open bash in the container

docker exec -it $(docker ps -a -q --filter "name=uflow-container") bash

Known issues

unknown flag: --gpus

Run the following command:

sudo apt-get install docker-ce-cli docker-ce

docker: Error response from daemon: could not select device driver "" with capabilities: [[gpu]].

Install NVIDIA GPU drivers. On Ubuntu use the Software&Updates UI, then run:

curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/ubuntu18.04/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
sudo apt-get update && sudo apt-get install -y nvidia-container-toolkit
sudo systemctl restart docker

libcudnn8 doesn't match when launching UFlow

Uninstall libcudnn8 with apt and follow the instructions at: https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#installlinux-deb