JavaCNN

A Java implementation of Convolutional Neural Network. This is a mavenized fork of https://github.com/BigPeng/JavaCNN refactored for the intention to use it in productive environments.

Original ideas are take from the DeepLearnToolbox (https://github.com/rasmusbergpalm/DeepLearnToolbox).

Include in your project

If you use maven, it's simple: Just add the dependency

<dependencies>

    <dependency>
        <groupId>javacnn</groupId>
        <artifactId>javacnn</artifactId>
        <version>0.4</version>
    </dependency>
   
</dependencies>

and ratopi's repository

<repositories>
    <repository>
        <id>ratopi.de releases</id>
        <url>http://ratopi.github.io/maven/releases/</url>
        <snapshots>
            <enabled>false</enabled>
        </snapshots>
    </repository>
</repositories>

to your project's pom.xml.

Build a CNN

LayerBuilder builder = new LayerBuilder();
builder.addLayer(Layer.buildInputLayer(new Size(28, 28)));
builder.addLayer(Layer.buildConvLayer(6, new Size(5, 5)));
builder.addLayer(Layer.buildSampLayer(new Size(2, 2)));
builder.addLayer(Layer.buildConvLayer(12, new Size(5, 5)));
builder.addLayer(Layer.buildSampLayer(new Size(2, 2)));
builder.addLayer(Layer.buildOutputLayer(10));
CNN cnn = new CNN(builder, 50);

Run on MNIST dataset

For running on MNIST dataset see project https://github.com/ratopi/javacnn.mnist.

Source Code

Get the source code from github:

git clone https://github.com/ratopi/JavaCNN.git 

License

MIT