Skip to content

Latest commit

 

History

History
125 lines (94 loc) · 3.22 KB

DOCKER.md

File metadata and controls

125 lines (94 loc) · 3.22 KB
Raw

Linux-Fake-Background-Webcam with Docker

Configuration

Docker-Compose will automatically read/create environment variables from an .env-file

So just copy the provided default-file and modify as needed:

cp defaults.env .env

To find the device number of your webcam you might consider v4l2-ctl:

sudo apt-get install v4l-utils
v4l2-ctl --list-devices

Docker Compose (CPU)

Prerequisites

Usage

  • Run and initial build containers: docker-compose up (or docker-compose up -d)
  • Stop and remove containers: docker-compose down
  • Note: Ctrl-C is currently stops the containers instead of changing images

Docker (GPU)

Note: The first time you start the GPU-version it will take a long time until the virtual cam can be used (around 5 min.). This is because CUDA does a lot of JIT-Caching when started for the first time. However, we can just use a volume for this cache. The subsequent start will be super fast. See: https://developer.nvidia.com/blog/cuda-pro-tip-understand-fat-binaries-jit-caching/

Prerequisites

  • v4l2loopback, see main readme
  • docker
  • docker-compose >= 1.27.0 (to support the runtime configuration in the docker-compose file)
  • Nvidia Docker

Using docker-compose

Works similar to the CPU-version, but you have to provide the docker-compose filename to use.

docker-compose -f docker-compose-gpu.yml up

When you stop the services make sure to delete the communication-volume (since the socket in this volume is currently not closed correctly):

docker-compose -f docker-compose-gpu.yml down --volumes
docker volume rm linux-fake-background-webcam_communication_volume

If you only want to switch the image (i.e. make changes in the .env file), you can just "restart" the fakecam-container:

docker-compose -f docker-compose-gpu.yml stop fakecam-gpu
docker-compose -f docker-compose-gpu.yml up -d fakecam-gpu

Manual via docker only

Build Images:

docker build -t bodypix -f ./bodypix/Dockerfile-gpu ./bodypix
docker build -t fakecam ./fakecam

Create a Network:

docker network create --driver bridge fakecam

Create a Volume:

docker volume create --name fakecam

Start the bodypix app with GPU support and listen on a UNIX socket:

docker run -d \
  --rm \
  --name=bodypix \
  --network=fakecam \
  -v fakecam:/socket \
  -e PORT=/socket/bodypix.sock \
  --gpus=all --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 \
  bodypix

Start the camera, note that we need to pass through video devices, and we want our user ID and group to have permission to them you may need to sudo groupadd $USER video:

docker run -d \
  --rm \
  --name=fakecam \
  --network=fakecam \
  --device=/dev/video2:/dev/video0 \
  --device=/dev/video11:/dev/video2 \
  -v fakecam:/socket \
  fakecam \
  -B /socket/bodypix.sock --no-foreground --scale-factor 1

After you've finished, clean up:

docker rm -f fakecam bodypix
docker volume rm fakecam
docker network rm fakecam