ailia Tech BLOG
ailia SDK Tutorial (C++)
This is a tutorial on how to use ailia SDK in C++. ailia SDK allows you to easily perform deep learning inference using C++ on the GPU.
C++ API Overview
The core of the ailia SDK is implemented in C++, which enables cross-platform execution of machine learning models using C++. Our SDK includes various samples making use of those APIs. This article describes how to build and run those samples.
License file setup
An license file is required to run ailia on Windows and Mac, but not on Linux, Jetson or RaspberryPi.
Place the file in the same folder as ailia.dll on Windows, and in the folder ~/Library/SHALO/ on Mac.
If you run the sample without a valid license file, an error AILIA_STATUS_LICENSE_NOT_FOUND = -20 will occur.
How to run samples on each platform
Mac
To use it on a Mac, both Xcode and the Xcode Command Line Tools are required. Xcode can be installed from the AppStore, the Command Line Tools using the following command.
xcode-select install
To download the machine learning models used in the samples, navigate to the samples/models folder and run the script download_model.sh
Here is what the what the folder looks like before and after downloading the models.
Next, copy the ailia core libraries library/mac/libailia.dylib and library/mac/libailia_blas.dylib to the folder samples/cpp
The samples use OpenCV for loading images and videos. It supports OpenCV 3 and 4, both can be installed using brew with the following command.
brew install opencv
The sample program expects OpenCV to be installed in /usr/local/lib, but depending on your environment, it may be installed in a different path and you may get the following error.
clang: error: no such file or directory: ‘/usr/local/lib/libopencv_core.dylib’
clang: error: no such file or directory: ‘/usr/local/lib/libopencv_imgproc.dylib’
clang: error: no such file or directory: ‘/usr/local/lib/libopencv_imgcodecs.dylib’
In that case, please use the following command to check the OpenCV path and rewrite the path in the Makefile.
find / -name=”opencv*”
For example, if OpenCV is found in /usr/local/Cellar/opencv/4.2.0_1, rewrite the contents of the Makefile as follows:
Then run the following command from the samples/cpp folder.
export OSTYPE=Mac
make
If successful, the folder content will look like this:
Run the object identification sample with the following command.
./ailia_classifier ../images/clock.jpg
Here is the result you should get.
Prototxt : ../models/keras-vgg16/VGG16.prototxt
Caffemodel : ../models/keras-vgg16/VGG16.caffemodel
ColorRange : 1
ColorFormat : 0
Prob : 0.773837 Class : analog clock
Prob : 0.179676 Class : wall clock
Prob : 0.030096 Class : stopwatch, stop watch
Prob : 0.005507 Class : barometer
Prob : 0.002026 Class : magnetic compass
Windows
On Windows, you need Visual Studio 2015 or newer and gnumake installed.
Visual Studio can be downloaded from the following link.
In order to install gnumake, download and run the executable labelled “Complete package, except sources” from the link below.
You need to add the gnumake executables to your environment variables (path), then you can run the make command from the command prompt to confirm that all is in place.
To download the machine learning model used in the sample, run the script download_model.bat in the folder samples/models.
Next, copy the dll and lib files from the folder library/windows/x64 to the sample folder samples/cpp.
On Windows, images are loaded via GDI+, therefore it is not necessary to install OpenCV.
Launch the “x64 Native Tools Command Prompt” associated to your version of Visual Studio and navigate to the samples/cpp folder.
Build the samples using the following command.
set OSTYPE=Windows
make
Run the object identification sample with the following command.
./ailia_classifier.exe ../images/clock.jpg
Linux
On Linux, use clang. clang is installed by default on Ubuntu 18.04LTS, the recommended environment for ailia SDK.
Since OpenCV and unzip are required to run the ailia SDK samples, install them using the following commands.
apt install libopencv-dev
apt install unzip
Run download_model.sh from the folder samples/models
./download_model.sh
Copy the .so library files from the folder library/linux to samples/cpp
Then run the following commands from the samples/cpp folder
export OSTYPE=Linux
make
Finally, run the object identification sample with the following command.
./ailia_classifier ../images/clock.jpg
iOS
In XCode, create a new iOS project and register libailia.a in the library/ios folder.
Add the frameworks Accelerate.framework, Metal.framework and MetalPerformanceShaders.framework to the build settings.
Since the ailia SDK is a C language interface, set the extension of the file that calls the ailia SDK to .mm in order to call the C language from ObjectiveC.
Trained models are registered in Supporting Files. The path to the registered trained models can be retrieved using pathForResource.
Thereafter, inferences can be made by calling the ailiaPredict API.
XCode project samples can be downloaded from the following link.
Android
For Android, the NDK is required, we specifically recommend NDK r19c. On Windows, Cygwin is also required.
https://developer.android.com/ndk/downloads?hl=en
To use the ailia SDK with the Android NDK, link libailia-VER-libc++_static.a in the library/android folder, where VER stands for the version of the library.
Make the following settings in Application.mk. The API LEVEL must be 16 or higher, and APP_STL must be set to c++_static.
Link the library to Android.mk by using the following settings.
The code to perform inference is written in main.cpp. The trained models must be copied to the SD card in advance.
The build will generate a .so file, which we will call from Java via JNI.
https://developer.android.com/ndk/samples/sample_hellojni
The ailia SDK also provides a standard JNI API, which can be called directly from Java. You can also use native-activity to create .apk with only C code.
https://developer.android.com/ndk/samples/sample_na
C++ samples
C++ samples can also be downloaded from the following repository.
ailia Inc. has developed ailia SDK, which enables cross-platform, GPU-based rapid inference.
ailia Inc. provides a wide range of services from consulting and model creation, to the development of AI-based applications and SDKs. Feel free to contact us for any inquiry.