Dockerfiles

Docker files for machine learning and robotics

Compiling

cd ~/dockerfiles/{pkg}
docker build --tag {tag} -f {file_name} .

Running

Command to get a bash (/bin/bash) from the container rmiyagusuku/dlbox:base-9.0 \ Container also has GPU capabilities (--runtime=nvidia)
and mounts the home directory to save data (-V /home:/home)

docker run --runtime=nvidia --rm -it \
    -v /home:/home \
    rmiyagusuku/dlbox:base-9.0 \
    /bin/bash

When running a hub version add the following options to allow the docker to use PAM authentication
also, if no command is specified (as bash in the previous example), the container automatically launches a hub at port 8000

docker run --runtime=nvidia --rm -it \
    --user $(id -u):$(id -g) \
    -v /etc/sudoers:/etc/sudoers:ro \
    -v /etc/pam.d:/etc/pam.d:ro \
    -v /etc/passwd:/etc/passwd:ro \
    -v /etc/shadow:/etc/shadow:ro \
    rmiyagusuku/dlbox:hub

For the ros versions it may be usefull to share the network with the host, so other ros processes can be launch from the host and still connect with the container (such as rviz for visualization)

docker run --runtime=nvidia --rm -it \
    --net=host \
    rmiyagusuku/ros:base

Most containers are not set up with users or allow running GUIs, yolo containers do
To enable GUI add DISPLAY and share /tmp/.X11-unix - more details in this post

docker run --runtime=nvidia --rm -it \
    -e DISPLAY=$DISPLAY \
    -v /tmp/.X11-unix:/tmp/.X11-unix \
    rmiyagusuku/yolov2:gpu