roboflow inference docker environment

containerized template to get up and running with a python installation containing the roboflow inference or inference-gpu module SDK.

features

• the build scripts and tools in this template configuration will use docker compose to spin up a debian-based docker container using the official python 3.10 bullseye docker image that has the inference/inference-gpu SDK installed via pip.

• build configuration options to install either inference or inference-gpu at the Makefile level.

• the docker compose specification is configured to allow pass-through for the /dev/video0 and /dev/video1 device files in order to enable the container to access the webcam and/or other camera devices bound to those files.

• the docker compose specification has the requisite volume mounts needed to allow display forwarding outside of the container and into the host via X11 in order to allow opencv to display GUI windows.

      - /tmp/.X11-unix:/tmp/.X11-unix
      - $HOME/.Xauthority:/root/.Xauthority

setup

make sure you have the following installed

• bash as the shell
• docker with the docker compose plugin
• (gnu) make

also make sure you have the X server running on the host

$ sudo X

create a .env file in the project's root and provide your Roboflow API key and model ID

ROBOFLOW_API_KEY=<YOUR_API_KEY_GOES_HERE>
ROBOFLOW_MODEL_ID=<YOUR_MODEL_ID_GOES_HERE>

run

type

$ make run

in order to

1. add +local:docker to the X server's access list
2. build the docker image with inference (CPU is the default COMPUTE_TYPE)
3. create the container
4. start the container
5. exec into the container via a bash session

alternatively, type

$ make run COMPUTE_TYPE='gpu'

in order to execute the same steps but build the docker image with inference-gpu instead if a GPU or any other kind of accelerator (OpenVINO, TensorRT) is available to be leveraged for faster inference

stop

first, make sure you've exited out the container by typing

$ exit

then, type

$ make stop

in order to

1. stop the container
2. remove the container
3. remove the built docker image

alternatively, type

$ docker compose down

if you just want to stop and remove the container but don't want to remove the built docker image (useful when you don't want to rebuild the entire image on each make start)

execing and exiting

after you've started the container for the first time, you can move between the host (local) environment and the running container environment seamlessly:

• you can exec into the already-running container again by just typing the same make start instruction mentioned before.
• you can exit out of the running container by just typing exit as mentioned before.