Introduction

2021-03-26 시점의 내용.

Apple Silicon M1로 빌드하자. OpenVINO도 Neural Compute Stick 2라면 움직입니다.

환경

Apple Silicon (M1) MacBook Air

Intel Compute Stick 2 (NCS2)

OpenVINO 2021.3 (No support CPU:arm)

OpenCV 4.5.3-openvino

MacPorts

sudo port install cmake gflags autoconf automake libtool libusb git-lfs

Miniforge3

conda install numpy matplotlib cython pandas scipy scikit-learn

build OpenCV 4.5.2-openvino

소스 트리

git clone --depth 1 -b 4.5.2-openvino https://github.com/opencv/opencv.git opencv-4.5.2-openvino

build

Install dir/opt/intel/openvino/opencv. Same as OpenVINO Distribution.

mkdir opencv-4.5.2-openvino/build
cd opencv-4.5.2-openvino/build

cmake \
    -DCMAKE_INSTALL_PREFIX=/opt/intel/openvino/opencv \
    -DCMAKE_BUILD_TYPE=Release \
    -DWITH_INF_ENGINE=ON \
    -DENABLE_CXX11=ON \
    -DWITH_TBB=ON \
    -DPYTHON3_EXECUTABLE=$HOME/miniforge3/bin/python3.9 \
    -DPYTHON3_LIBRARY=$HOME/miniforge3/lib/libpython3.9.dylib \
    -DPYTHON3_INCLUDE_DIR=$HOME/miniforge3/include/python3.9 ..

make -j 4
sudo make install

build OpenVINO 2021.3

소스 트리

git clone -b 2021.2 https://github.com/openvinotoolkit/openvino.git openvino-2021.3
cd openvino-2021.3
git submodule update --init --recursive

build

Install dir/opt/intel/openvino. Same as OpenVINO Distribution.

mkdir openvino-2021.3/build
cd openvino-2021.3/build

cmake \
    -DCMAKE_INSTALL_PREFIX=/opt/intel/openvino \
    -DCMAKE_BUILD_TYPE=Release \
    -DENABLE_PYTHON=ON \
    -DPYTHON_EXECUTABLE=$HOME/miniforge3/bin/python3.9 \
    -DPYTHON_LIBRARY=$HOME/miniforge3/lib/libpython3.9.dylib \
    -DPYTHON_INCLUDE_DIR=$HOME/miniforge3/include/python3.9 ..

make -j 4
sudo make install

patch this.

*** setupvars.sh.orig   2020-12-17 12:28:59.000000000 +0900
--- setupvars.sh    2020-12-17 12:29:34.000000000 +0900
***************
*** 18,23 ****
--- 18,24 ----
  BASE_DIR="$( dirname "$SCRIPT_DIR" )"

  INSTALLDIR="${BASE_DIR}"
+ INSTALLDIR=/opt/intel/openvino

environment valiable

source /opt/intel/openvino/bin/setupvars.sh
export ngraph_DIR=/opt/intel/openvino/deployment_tools/ngraph/cmake
export OpenCV_DIR=/opt/intel/openvino/opencv 
export PYTHONPATH=$OpenCV_DIR/lib/python3.9/site-packages:$PYTHONPATH

※ 서로 의존하고 있으므로 OpenCV → OpenVINO → OpenCV에서 빌드&설치하는 것.

Run with Intel Neural Compurte Stick 2

python opencv-4.5.2-openvino/samples/python/video_threaded.py

python /opt/arm64/openvino/deployment_tools/inference_engine/samples/python/classification_sample_async/classification_sample_async.py -m alexnet.xml -d MYRIAD -i image.jpg

OpenCV DNN backend Inference Engine (OpenVINO: Python, C++)

Run Success!! 파이썬에서도 작동합니다.

Reference

(최소 스텝 밖에 쓰고 있지 않기 때문에) 여기에 써 있는 채로 잘 가지 않거나, 환경을 다소 변경하고 싶은 경우는 이하에 눈을 통하면 할 수 있지 않을까.

M1 MacBook에서 개발 환경 구축

OpenVINO wiki: BuildingForMacOS

TODO

CoreML도 시험하고 싶다, , , 하지만. OpenCV ML backend에서 CoreML을 지원하지 않습니까?