SSH and X11 forwarding enabled OpenVINO Docker image

Summary

• Provides Dockerfile for OpenVINO developers
• Enable SSH for easy development
• Enable X11 redirect to display graphics on the host environment
• Setup some useful aliases for easy access to the OpenVINO tools
• Setup and start OpenVINO development in 5 min

• Added Dockerfile for OpenVINO 2022.1. (./dockerfile_openvino2022.1/Dockerfile)

Description

Intel provides several types of OpenVINO Docker images but those images doesn't enable SSH. This project provides a Dockerfile which enables SSH. Developers can connect to the container with SSH client, and use X11 redirect to display graphics on the host machine.

How to build a Docker image

docker build -t openvino_dev_ssh .

How to start a Docker container

Container will start and resides as a background process (daemon). SSH server will up and run, and wait for remote connection on port 22.

docker run -d --rm -p 22:22 --name ov -it openvino_dev_ssh

Note: To enable X11 redirect on Ubuntu system, user needs to add -e DISPLAY=${DISPLAY} -v /tmp/.X11-unix/:/tmp/.X11-unix/ options. Refer to 'How to enable X11 redirect (Host=Ubuntu)' section for details.

Linux users only

• Enable integrated GPU.

docker run -d --rm -p 22:22 --name ov -it --device=/dev/dri openvino_dev_ssh

• Enable VPU (Myriad-X, Neural Compute Stick 2:NCS2)

docker run -d --rm -p 22:22 --name ov -it --device=/dev/ion -v /var/tmp:/var/tmp openvino_dev_ssh

• Please refer to the OpenVINO official web documentation for details.
  Run the Docker* Image for GPU
  Use a Docker* Image for Intel® Neural Compute Stick 2

How to connect to the container with ssh

You can use any ssh client program to connect to the container.
Note: password is 'ovuser'.

ssh ovuser@localhost

How to stop the container

Caution: The contents in the container will be lost when you stop the container if you put '--rm' option at the time when you start the container.

docker stop ov

How to enable X11 redirect (Host=Windows)

Install Xserver software

Win10/11 users need to install and start a X server program such as VcXsrv.

Disable access control in Xserver on host

Xservers denies remote access in default. You need to disable this access control to make the Xserver accepts the drawing request from remote programs.
Windows user must refer to the Xserver document and disable the access control.
'VcXsrv' has Disable access control check box in the dialog box which appears when you start the XLaunch app.

Set DISPLAY environment in the container to redirect X11 drawing to the host.

(container) export DISPLAY=192.168.1.21:0.0
Note: 192.168.1.21 is the host IP address

• Now you can run the X11 apps in the container

How to enable X11 redirect (Host=Ubuntu)

Disable access control in Xserver on host

Xservers denies remote access in default. You need to disable this access control to make the Xserver accepts the drawing request from remote programs.

xhost +

Share host X authentication info and DISPLAY environment variable

Add -e DISPLAY=${DISPLAY} -v /tmp/.X11-unix/:/tmp/.X11-unix/ to docker run command line to share the host X authentication information with the container.

docker run -d --rm -p 22:22 --name ov -it -e DISPLAY=${DISPLAY} -v /tmp/.X11-unix/:/tmp/.X11-unix/ openvino_dev_ssh

• Now you can run the X11 apps in the container

Building OpenVINO sample apps in the container

• Build C++ sample apps
  Built binaries can be found in ~/inference_engine_cpp_samples_build/intel64/Release/

cd ${INTEL_OPENVINO_DIR}/deployment_tools/inference_engine/samples/cpp/
./build_samples.sh 

• Build C sample apps
  Built binaries can be found in ~/inference_engine_c_samples_build/intel64/Release/

cd ${INTEL_OPENVINO_DIR}/deployment_tools/inference_engine/samples/c/
./build_samples.sh 

• Copy Python demos to the user directory

cp -r ${INTEL_OPENVINO_DIR}/deployment_tools/inference_engine/samples/python/ ~/python_demos

Running OpenVINO sample apps

• Image classification sample app

Copy input image files.

cd ~
cp $INTEL_OPENVINO_DIR/deployment_tools/demo/car.png .
cp $INTEL_OPENVINO_DIR/deployment_tools/demo/car_1.bmp .

Download a DL model with model downloader.

omz_downloader --name googlenet-v1-tf
omz_converter --name googlenet-v1-tf --precisions FP16

Run a C++ image classification sample app.

inference_engine_cpp_samples_build/intel64/Release/classification_sample_async -m public/googlenet-v1-tf/FP16/googlenet-v1-tf.xml -i car.png 

Run a Python image classification sample app.

python3 python_demos/hello_classification/hello_classification.py -m public/googlenet-v1-tf/FP16/googlenet-v1-tf.xml -i car.png 

Using webCam from the apps in the container (Linux only)

Linux can share USB webCam device with the container. Add --device=/dev/video0 to docker run command line.

Others

The Dockerfile setup some aliases to the OpenVINO tools for developers convenience. Developers can access to the tools without remembering the long-and-deep path to the command.

alias/command	tool	description
omz_downloader	OMZ* model downloader	Download OMZ models by specifying the model name
omz_converter	OMZ* model converter	Convert downloaded OMZ models in framework format into OpenVINO IR model format
omz_quantizer	OMZ* model quantizer	Quantize OMZ IR models into INT8 precision
omz_info_dumper	OMZ* model info dumper	Display OMZ model information
benchmark_app	Benchmark app	Run IR model benchmark with various options
mo.py	Model Optimizer	Convert and optimize DL models in framework format into OpenVINO IR format
accuracy_check	Accuracy Checker	Check inference accuracy of an OpenVINO IR model
pot	Post-training Optimization Tool	Quantize IR model into INT8

Note: OMZ = Intel Open Model Zoo