在 Raspebrry pi 4 上加载 openvino python 库

Question

After followign the guides (and coming through the github trackers), I was able to get OpenVINO to install on my pi4 and can run /opt/intel/openvino/bin/armv7l/Release/object_detection_sample_ssd successfully using my own trained model.

I used the follow cmake command most recently, but I've done pretty much all iterations I could find on the several instruction pages.

cmake -DCMAKE_BUILD_TYPE=Release /
      -DENABLE_SSE42=OFF /
      -DTHREADING=SEQ /
      -DENABLE_GNA=OFF /
      -DENABLE_PYTHON=ON /
      -DPYTHON_EXECUTABLE=/usr/bin/python3 /
      -DPYTHON_LIBRARY=/usr/lib/arm-linux-gnueabihf/libpython3.7m.so /
      -DPYTHON_INCLUDE_DIR=/usr/include/python3.7m /
      -NGRAPH_PYTHON_BUILD_ENABLE=ON /
      -DCMAKE_CXX_FLAGS=-latomic /
      -DOPENCV_EXTRA_EXE_LINKER_FLAGS=-latomic /
      -D CMAKE_INSTALL_PREFIX=/usr/local  /
      .. && make

When I try one of the python examples I get

ModuleNotFoundError: No module named 'ngraph'

Looking more into it now and it appears the issue is the "setupvars.sh" script not calling into the right directory. I was able to get the module openvino to load by adjusting the export path. I must say the amount of documentation that is, quite frankly all over the place and seems to have wrong directory structures left and right.

Answer 1

As of right now, the builds for the Raspberry Pi are not being made with NGRAPH compiled. The user has to compile it themselves from scratch per this thread. https://community.intel.com/t5/Intel-Distribution-of-OpenVINO/No-ngraph-bindings-for-python-in-raspbian-distribution/mp/1263401#M23093

The comment by the Intel "Support" is wrong and won't work on new builds.

Answer 2

Please refer to the steps below for building Open Source OpenVINO™ Toolkit for Raspbian OS:

1. Set up build environment and install build tools

   sudo apt update && sudo apt upgrade -y

   sudo apt install build-essential

2. Install CMake from source

   cd ~/

   wget https://github.com/Kitware/CMake/releases/download/v3.14.4/cmake-3.14.4.tar.gz

   tar xvzf cmake-3.14.4.tar.gz

   cd ~/cmake-3.14.4

   ./bootstrap

   make -j4 && sudo make install

3. Install OpenCV from source

   sudo apt install git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev python3-scipy libatlas-base-dev

   cd ~/

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

   cd opencv && mkdir build && cd build

   cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr/local..

   make -j4 && sudo make install

4. Download source code and install dependencies

   cd ~/

   git clone --depth 1 --branch 2021.3 https://github.com/openvinotoolkit/openvino.git

   cd ~/openvino

   git submodule update --init --recursive

   sh./install_build_dependencies.sh

   cd ~/openvino/inference-engine/ie_bridges/python/

   pip3 install -r requirements.txt

5. Start CMake build

   export OpenCV_DIR=/usr/local/lib/cmake/opencv4

   cd ~/openvino

   mkdir build && cd build

   cmake -DCMAKE_BUILD_TYPE=Release \

   -DCMAKE_INSTALL_PREFIX=/home/pi/openvino_dist \

   -DENABLE_MKL_DNN=OFF \

   -DENABLE_CLDNN=OFF \

   -DENABLE_GNA=OFF \

   -DENABLE_SSE42=OFF \

   -DTHREADING=SEQ \

   -DENABLE_OPENCV=OFF \

   -DNGRAPH_PYTHON_BUILD_ENABLE=ON \

   -DNGRAPH_ONNX_IMPORT_ENABLE=ON \

   -DENABLE_PYTHON=ON \

   -DPYTHON_EXECUTABLE=$(which python3.7) \

   -DPYTHON_LIBRARY=/usr/lib/arm-linux-gnueabihf/libpython3.7m.so \

   -DPYTHON_INCLUDE_DIR=/usr/include/python3.7 \

   -DCMAKE_CXX_FLAGS=-latomic..

   make -j4 && sudo make install

6. Configure the Intel® Neural Compute Stick 2 Linux USB Driver

   sudo usermod -a -G users "$(whoami)"

   source /home/pi/openvino_dist/bin/setupvars.sh

   sh /home/pi/openvino_dist/install_dependencies/install_NCS_udev_rules.sh

7. Verify nGraph module binding to Python

   cd /home/pi/openvino_dist/deployment_tools/inference_engine/samples/python/object_detection_sample_ssd

   python3 object_detection_sample_ssd.py -h