A simple neural network that recognise handwritten characters
A demo of the application can be found here: https://www.youtube.com/watch?v=Zn5UzdW73OA
The neural network is written in C++. Instead, the GUI and the application of the neural network is written in Java (that involves loading MNINST images, scaling and converting them to an usable format and so on...) . C++ neural network functionalities are accessed from Java through JNI.
Currently there are pre-compiled natives only for Windows x64 and Linux x64 that will be loaded dinamically when starting the Java application.
If you need to run this on other SO you have to compile the C++ source to get an external library file (.so on linux, .dll on windows etc...) that you need to load into the java NeuralNetwork class. Search about Java JNI to get more info on how to link native libraries to java code.
Training/Test data used in the above video can be found here: http://yann.lecun.com/exdb/mnist/
Pre-compiled versions can be found in Releases.
You can build the application by using gradle. Java JDK min 1.8 is necessary to build this project. Open the terminal and move to the root of this project then execute
gradlew assembleBuild files will be generated into a folder named build. The executable .jar will be generated into build/libs.
This version comes with pre-compiled Windows x64 and Linux x64 natives. If you want this application to run on a diffent OS you need to compile the C++ sources.
Source files are located into src/main/cpp. You will need CMake to compile the library.