NullPointer Exception on initialization of BasicMLDataSet with Input and Ideal Double 2d Arrays

Question

I am currently working on a CNN that can tell the difference between a dog and a cat. This is the code I have worked out so far:

public bool trained = false;
public BasicNetwork network;
public Form1()
{
  InitializeComponent();
}
[..]
private void trainNN()
{    
  double[][] input = LoadImages();
  double[][] ideal = LoadIdeal();
  var trainingSet = new BasicMLDataSet(input, ideal);
  var train = new ResilientPropagation(network, trainingSet);
  network = CreateNetwork();
}
    private double[][] LoadImages()
    {
        status.Text = "Loading images...";
        String[] dogimgs = Directory.GetFiles(Directory.GetCurrentDirectory() + "\\dog_img\\", "*", SearchOption.TopDirectoryOnly);
        String[] catimgs = Directory.GetFiles(Directory.GetCurrentDirectory() + "\\cat_img\\", "*", SearchOption.TopDirectoryOnly);
        int dogimgscount = dogimgs.Length;
        int catimgscount = catimgs.Length;
        int totalimgscount = dogimgscount + catimgscount;
        double[][] images = new double[totalimgscount][];
        for (int dogloop = 0; dogloop < dogimgscount; dogloop++)
        {
            status.Text = "Loading images... [" + (dogloop + 1) + "/" + totalimgscount + "]";
            images[dogloop] = Image2Matrix(new Bitmap(dogimgs[dogloop]));
        }
        for (int catloop = 0; catloop < catimgscount; catloop++)
        {
            status.Text = "Loading images... [" + (catloop + dogimgscount) + "/" + totalimgscount + "]";
            images[catloop + dogimgscount - 1] = Image2Matrix(new Bitmap(catimgs[catloop]));
        }
        status.Text = "Images loaded.";
        return images;
    }
    private double[][] LoadIdeal()
    {
        String[] dogimgs = Directory.GetFiles(Directory.GetCurrentDirectory() + "\\dog_img\\", "*", SearchOption.TopDirectoryOnly);
        String[] catimgs = Directory.GetFiles(Directory.GetCurrentDirectory() + "\\cat_img\\", "*", SearchOption.TopDirectoryOnly);
        int dogimgscount = dogimgs.Length;
        int catimgscount = catimgs.Length;
        int totalimgscount = dogimgscount + catimgscount;
        double[][] ideal = new double[totalimgscount][];
        for (int dogloop = 0; dogloop < dogimgscount; dogloop++)
        {
            ideal[dogloop] = new[] { 0.0, 1.0 };
        }
        for (int catloop = 0; catloop < catimgscount; catloop++)
        {
            ideal[catloop + dogimgscount - 1] = new[] { 1.0, 0.0 };
        }
        return ideal;
    }

I know this might not be the smartest way to load the images, but I just want to see the concept work before I start boosting performance. My problem is the following: If I put 4 images, 2 in dog_img, 2 in cat_img, the program loads the images fine and both the input and the ideal array have a length of 4 and they both are filled with double values. But on the line

var trainingSet = new BasicMLDataSet(input, ideal);

the program throws a NullPointerException Error. Both arrays are clearly initialised and not null or empty, yet it still throws the error. Any help is appreciated. Thank you. FritzFurtz

Answer 1

Encog does not support CNN's, all input to Encog neural networks are 1D vectors. So, if you want to feed an image to Encog, you need to squash it to a 1D vector and wrap it in a BasicMLData object. Encog is probably not the best solution for computer vision/CNN. For C#, I would look at CNTK .