Build custom Intel® Distribution of OpenVINO™ toolkit Docker image

This repository folder contains the Dockerfile to build a docker image with the Intel® Distribution of OpenVINO™ toolkit.

Components

• OpenVINO™ Toolkit - Deep Learning Deployment Toolkit repository
• OpenCV: Open Source Computer Vision Library
• OpenVINO™ Toolkit - Open Model Zoo repository

How to build

Go to the folder with the Dockerfile and run:

docker build -t [image:tag] .

• /opt/intel/openvino folder will contain OpenVINO build
• /opt/intel/repo will contain OpenVINO™ and OpenCV git repositories in openvino and opencv folders accordingly.

If you want to rebuild the entire image, use the docker --no-cache option:

docker build --no-cache -t image:tag .

You can use the docker --build-arg option to override the following variables:

• OPENVINO_FORK - To specify a GitHub fork of the OpenVINO repository to use. By default, it is main OpenVINO repository.
• OPENVINO_BRANCH, OPENCV_BRANCH, OMZ_BRANCH - To specify branches with source code. By default, they are equal to "master".
• BUILD_OPENCV_CONTRIB - If set to yes, OpenCV will be built with extra modules (default is no). You can use the OPENCV_CONTRIB_BRANCH argument to specify a branch in the opencv_contrib repository.
• OCL_VERSION - To specify the version of Intel® Graphics Compute Runtime for OpenCL™ Driver on Linux. By default, it is equal to "19.41.14441".

For example:
This command builds an image with OpenVINO™ 2021.2 release.

docker build -t openvino:2021.2 --build-arg OPENVINO_BRANCH="releases/2021/2" .

You can manually set up cmake parameters to build a custom package from source code using these files:

• openvino_cmake.txt
• opencv_cmake.txt

Note:
By default, these files already contain some parameters for Debug build
Do not override PATH/PREFIX options. This can break a build of package.

Build stages

The docker image is built using a multi-step build:

1. setup_openvino
   Clone OpenVINO™ git repository with submodules and install build dependencies.
   Open Model Zoo will be included as a submodule of OpenVINO.
2. build_openvino
   Build OpenVINO™ (CPU, iGPU, VPU support) with the parameters specified in openvino_cmake.txt.
   It does not include OpenCV.
3. copy_openvino
   Copy OpenVINO™ build to clear Ubuntu:20.04 image.
4. openvino
   Install OpenVINO™ dependencies. Now you can use it.
5. opencv
   Build and setup OpenCV with the parameters specified in opencv_cmake.txt.
   OpenCV can be optionally built with extra modules (see the BUILD_OPENCV_CONTRIB argument description above).

Use the docker --target option to specify a final stage.

docker build --target [stage] -t [image:tag] .

For example:
This command builds an image without OpenCV.

docker build --target openvino -t ie:latest .

How to test

You can use our default pipeline to test your image:

python3 docker_openvino.py test -t [image:tag] -dist custom 

Note:
Docker CI framework automatically runs the corresponding tests.
By default, the product version is equal to the latest release version. Use -p to override this.

How to run

Please follow the Run built image section in Docker CI getting started guide.

Prebuilt images

Prebuilt images are available on:

• Docker Hub
• Red Hat* Quay.io
• Red Hat* Ecosystem Catalog

License

This Dockerfile contains third-party components with different licenses.
If you are distributing the container as a whole, then you are responsible for license compliance for all of the software it contains.

Documentation

• Install Intel® Distribution of OpenVINO™ toolkit for Linux* from a Docker* Image
• Install Intel® Distribution of OpenVINO™ toolkit for Windows* from Docker* Image
• Official Dockerfile reference

---

* Other names and brands may be claimed as the property of others.