NCS2-and-Coral-TPU-on-aarch64

KO | EN

gitlite — search

#python #ai #llm #machine-learning #deep-learning #linux #docker #awesome #kubernetes #awesome-list #pytorch #security #openai #ios

NCS2-and-Coral-TPU-on-aarch64

(★ 8)

How to install NCS2 Movidius X and Google Coral TPU on aarch64 (ARM64)

NCS2-and-Coral-TPU-on-aarch64 Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

**How to install NCS2 Myriad X and Google Coral TPU on aarch64 (ARM64)**

This is tested with :

- Nvidia Jetson Nano ( Ubuntu 18.04 )
 - NanoPi M4 & NanoPi Neo4 ( Armbian 18.04 )
 
 Coral just needs a minor fix, step 6 
 OpenCV step 5, is optional if you already have it installed (Nvidia Jetson Nano) 
 Step 7-8 is aimed at NCS2, and builds all C++ dldt inference-engine libraries and ie_api.so for python

**1. apt-get** 
 
```bash
$sudo apt-get update && sudo apt-get upgrade
$sudo apt-get install build-essential cmake unzip pkg-config
$sudo apt-get install libavcodec-dev libavformat-dev libswscale-dev
$sudo apt-get install libxvidcore-dev libx264-dev 
$sudo apt-get install libgtk-3-dev libopenblas-dev libpng-dev
$sudo apt-get install libatlas-base-dev gfortran libdc1394-22-dev
$sudo apt-get install python3-dev python3-numpy python3-pip
$sudo apt-get install curl wget libssl-dev ca-certificates git libboost-regex-dev
$sudo apt-get install libgtk2.0-dev pkg-config unzip automake libtool autoconf
$sudo apt-get install libcairo2-dev libpango1.0-dev libglib2.0-dev libgtk2.0-dev
$sudo apt-get install libgstreamer1.0-0 gstreamer1.0-plugins-base libusb-1.0-0-dev
$sudo apt-get install ffmpeg libjpeg-dev libtiff-dev git libtbb2 libtbb-dev
$sudo apt-get install cython cython3 libv4l-dev
```

**2. pip/python** 
```bash
$wget https://bootstrap.pypa.io/get-pip.py
$sudo python3 get-pip.py
$sudo -H pip install numpy
$sudo -H pip3 install Cython
```

**3. Extra nice to have** 
```bash
$sudo apt-get install codeblocks codeblocks-contrib
$sudo apt-get install spyder3
```

**4. Swapfile Armbian** if you want to compile faster, with make -j4 
 https://linuxize.com/post/how-to-add-swap-space-on-debian-9/ 
```bash
$sudo swapon --show
$sudo fallocate -l 2G /swapfile
$sudo dd if=/dev/zero of=/swapfile bs=1024 count=2048576
$sudo chmod 600 /swapfile
$sudo mkswap /swapfile
$sudo swapon /swapfile
$# Add this to make it permanent
$sudo nano /etc/fstab
/swapfile swap swap defaults 0 0 
```
**5. OpenCV, select your prefered version** 
This step is optional ( intended primary for a clean Armbian ) 
```bash
$cd ~
$wget -O opencv_contrib.zip https://github.com/opencv/opencv_contrib/archive/4.0.1.zip
$wget -O opencv.zip https://github.com/opencv/opencv/archive/4.0.1.zip
$unzip opencv_contrib.zip 
$unzip opencv.zip
$mv opencv-4.0.1 opencv
$mv opencv_contrib-4.0.1 opencv_contrib
$cd opencv
$mkdir build
$cd build
$#Todo add more flags of your choice....and note opencv directory
$cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D INSTALL_PYTHON_EXAMPLES=ON -D INSTALL_C_EXAMPLES=OFF -D OPENCV_ENABLE_NONFREE=ON -D OPENCV_EXTRA_MODULES_PATH=~/opencv_contrib/modules -D BUILD_EXAMPLES=ON -D ENABLE_PRECOMPILED_HEADERS=OFF -DWITH_INF_ENGINE=ON -DENABLE_CXX11=ON ..
$make -j4
$sudo make install
```
 
 
**6. Coral** 
 https://coral.withgoogle.com/docs/accelerator/get-started/ 
```bash
$cd ~ 
$mkdir coral 
$cd coral 
$wget https://dl.google.com/coral/edgetpu_api/edgetpu_api_latest.tar.gz -O edgetpu_api.tar.gz --trust-server-names 
$tar xzf edgetpu_api.tar.gz 
$cd edgetpu_api 
$bash ./install.sh 
```
 
Minor simple stupid fix for Coral

```bash
$cd /usr/local/lib/python3.6/dist-packages/edgetpu/swig/
$sudo cp _edgetpu_cpp_wrapper.cpython-35m-aarch64-linux-gnu.so edgetpu_cpp_wrapper.so
```

**7. NCS 2** 
https://software.intel.com/en-us/articles/ARM64-sbc-and-NCS2 
https://github.com/skhameneh/OpenVINO-ARM64 
```bash
$cd ~
$mkdir ncs2
$cd ncs2
$git clone https://github.com/opencv/dldt.git
$cd dldt/inference-engine/
$git submodule init
$git submodule update --recursive
```

**7.1 Prepare dldt for aarch64 or architecture of your choice** 
 *Check your machine hardware architecture with : ```uname --m``` in my case aarch64* 
 - Add more samples than comes with dldt
 - Fix #define in samples !defined(__aarch64__)
 - Add aarch64 build folder ( optional )
 - Select compiler ( optional )
 - Select proper python include & library
 - patch output folder with tool from x86 installation
 - install python files

**7.1.1 Add more samples than comes with dldt** 
 
from 
l_openvino_toolkit_raspbi_p_2019.1.094.tgz 
to 
```bash
$~/ncs2/dldt/inference-engine/samples/
```

**7.1.2 Fix defines in samples** 
*Ex, with codeblocks Ctrl+Shift+F ( find in files )* 
find ```!defined(_M_ARM)``` then add ```&& !defined(__aarch64__)```
 
example : ```#if !defined(__arm__) && !defined(_M_ARM) && !defined(__aarch64__)```

**7.1.3 Add aarch64 build folder** ( optional ) 
 
```
find "armv7l" and add elseif(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64")

# example 
if(CMAKE_SYSTEM_PROCESSOR STREQUAL "armv7l") 
    set (ARCH_FOLDER armv7l) 
elseif(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64") 
    set (ARCH_FOLDER aarch64) 
```

**7.1.4 Select compiler** ( optional ) 
 
Find and update ```CMAKE_C_COMPILER``` 
```
set(CMAKE_SYSTEM_NAME Linux)
set(CMAKE_SYSTEM_PROCESSOR aarch64)
set(CMAKE_C_COMPILER gcc-7)
set(CMAKE_CXX_COMPILER g++-7) 
```

**7.2 Build inference-engine for C++ & python** 
 
*from step 5, depends on your version of OpenCV*

```bash
$export OpenCV_DIR=/usr/include/opencv2 # Nvidia Jetson nano 
```
or 
```bash
$export OpenCV_DIR=/usr/local/include/opencv4 # Install from source 
```
 
verify location for cmake( adjust if needed)

``` 
-DPYTHON_LIBRARY=/usr/lib/aarch64-linux-gnu/libpython3.6m.so 
-DPYTHON_INCLUDE_DIR=/usr/include/python3.6 
 ```

```bash
$cd ~/ncs2/dldt/inference-engine
$mkdir build
$cd build
$ #Todo add more flags of your choice... 
$cmake -DENABLE_PYTHON=ON -DPYTHON_EXECUTABLE=`which python3.6` -DPYTHON_LIBRARY=/usr/lib/aarch64-linux-gnu/libpython3.6m.so -DPYTHON_INCLUDE_DIR=/usr/include/python3.6 -DCMAKE_BUILD_TYPE=Release -DENABLE_MKL_DNN=OFF -DENABLE_CLDNN=OFF -DENABLE_GNA=OFF -DENABLE_SSE42=OFF -DTHREADING=SEQ ..
$make -j4
```

**7.3 create tools folder** 
patch output folder with tool from x86 installation

```
~/ncs2/dldt/inference-engine/bin/aarch64/Release/lib/python_api/python3.6/openvino/inference_engine/tools
```

copy from your x86 openvino tools folder to tools

```bash
tools/accuracy_checker
tools/benchmark
tools/calibration
tools/utils
tools/network.py
```

**7.4 Simple stupid python installation**

```bash
$cd /usr/local/lib/python3.6/dist-packages 
$sudo mkdir openvino 
$cd openvino 
$sudo cp -r ~/ncs2/dldt/inference-engine/bin/aarch64/Release/lib/python_api/python3.6/openvino/* . 
```

**8.0 Myriad-rules ?** ( seems to work without it ) 
*from l_openvino_toolkit_raspbi_p_2019.1.094.tgz*

```bash
$sudo cp 97-myriad-usbboot.rules_.txt /etc/udev/rules.d/97-myriad-usbboot.rules
$udevadm control --reload-rules
$udevadm trigger
$ldconfig 
```

**Ready !** 
 
Test a sample ( if it doesn't start ..try use sudo python3 )

NCS2-and-Coral-TPU-on-aarch64

NCS2-and-Coral-TPU-on-aarch64 Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

NCS2-and-Coral-TPU-on-aarch64

NCS2-and-Coral-TPU-on-aarch64 Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README