Skip to content

Containerized OpenVINO Toolkit (+OpenCV) for arm64v8-based SBCs

Notifications You must be signed in to change notification settings

likeablob/openvino-opencv-docker

Repository files navigation

openvino-opencv-docker

Containerized OpenVINO Toolkit (+OpenCV) to use Intel Movidius Neural Compute Stick especially on arm64v8 based SBCs (such as Orange Pi + Armbian).
This might be also useful if you want just OpenCV4 w/o any HW DNN accelerator.

Images

All the images are based on debian:buster. See also Getting started.

  image: registry.gitlab.com/likeablob/openvino-opencv-docker:master-runtime-arm64v8
  image: registry.gitlab.com/likeablob/openvino-opencv-docker:master-runtime-x86_64

Tagging rule

  • ${branch}-${type}-${arch}
    • ${branch}: A branch name. master or devel.
    • ${type}: runtime or build.
      • runtime: A image containing the OpenVINO and OpenCV runtime libs.
      • build: A image to build OpenVINO and OpenCV.
    • ${arch}: arm64v8 or x86_64. (The arm64v8 images can be run on x86_64 system thanks to qemu-aarch64-static.)

Getting started

A simple face detection demo written in Python with real-time visualization on the display.
The DNN model used is opencv/open_model_zoo/face-detection-retail-0004.

git clone https://github.com/likeablob/openvino-opencv-docker
cd openvino-opencv-docker/demo-face-detection-x11
./download_model.sh
T_ARCH="arm64v8" docker-compose run --rm app
T_ARCH="x86_64" docker-compose run --rm app

A simple face detection demo written in Python. Utilize a ws2812 LED strip to visualize the result. (ws2812 is controlled via SPI bus).
The DNN model used is opencv/open_model_zoo/face-detection-retail-0004.

(*Note that currently this demo can only be run on arm64v8 systems.)

git clone https://github.com/likeablob/openvino-opencv-docker
cd openvino-opencv-docker/demo-face-detection-ws2812
./download_model.sh
docker-compose run --rm app
# CAM: WxH=640.0x480.0 FPS=30.0
# ...
# top_result: [0.         1.         0.7441406  0.32885742 0.5986328  0.4699707 0.8642578 ]
# inference_fps(mean):14.38, inference_fps: 25.38, total_fps: 19.93
# ...

Appendix 1: Installing Docker on Armbian

curl -fsSL https://get.docker.com -o get-docker.sh
sudo sh ./get-docker.sh
sudo usermod -aG docker $USER

sudo apt install -y gnupg2 pass docker-compose

Appendix 2: Enabling SPI on Armbian

sudo vi /boot/armbianEnv.txt
+ overlays=spi-spidev
+ param_spidev_spi_bus=0 # param_spidev_spi_bus=1 for OPi-Zero*
sudo shutdown -r now

ls /dev/spidev*
# /dev/spidev0.0
# `docker-compose run --rm app -b 1 0` to use /dev/spidev1.0
Orange Pi Lite2 WS2812
Pin 19(SPI0_MOSI) DI
Pin 25(GND) GND
Pin 1(3.3V) 3.3V

About

Containerized OpenVINO Toolkit (+OpenCV) for arm64v8-based SBCs

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published