schuler/pascal-cai-neural

Pre-trained Neural API Networks (Models)

This repository contains pre-trained neural network models for the CAI neural API.

Super resolution pre-trained neural network model

You can icrease the resolution of your own images with this code and its pre-trained model. After compiling the super resolution code, you will be able to increase the resolution of your own images via command line:

#SuperResolution -i street.png -o street2.png

The parameter -i defines the input file while -o defines the output file. You can find more details at this link.

Image classification pre-trained neural network models

Dataset	Source Code	Input Size	Trained Model	Parameters	Test Accuracy
Malaria	source	64x64x3	Malaria-20230720	192K	95.63%
Colorectal Cancer	source	64x64x3	Colorectal-20230720	202K	94.26%
Plant Leaf Disease (Plant Village)	source	64x64x3	SimplePlantLeafDisease-20230720	252K	99.03%

Using Trained Models for Image Classification

The simplest way to load a trained model and classify an image is:

  procedure ClassifyOneImageSimple;
  var
    NN: TNNet;
    ImageFileName: string;
    NeuralFit: TNeuralImageFit;
  begin
    WriteLn('Loading Neural Network...');
    NN := TNNet.Create;
    NN.LoadFromFile('SimplePlantLeafDisease-20230720.nn');
    NeuralFit := TNeuralImageFit.Create;
    ImageFileName := 'plant/Apple___Black_rot/image (1).JPG';
    WriteLn('Processing image: ', ImageFileName);
    WriteLn(
      'The class of the image is: ',
      NeuralFit.ClassifyImageFromFile(NN, ImageFileName)
    );
    NeuralFit.Free;
    NN.Free;
  end;  

The above source code is located at TestPlantLeafDiseaseTrainedModelOneImage.pas.

If you would like to test against the actual training dataset, you can follow this example: TestPlantLeafDiseaseTrainedModel.pas.

In the case that you need more control on how your image is classified, you can look at this more detailed example:

  procedure ClassifyOneImage;
  var
    NN: TNNet;
    ImageFileName: string;
    NeuralFit: TNeuralImageFit;
    vInputImage, vOutput: TNNetVolume;
    InputSizeX, InputSizeY, NumberOfClasses: integer;
  begin
    WriteLn('Loading Neural Network...');
    NN := TNNet.Create;
    NN.LoadFromFile('SimplePlantLeafDisease-20230720.nn');
    NN.DebugStructure();
    InputSizeX := NN.Layers[0].Output.SizeX;
    InputSizeY := NN.Layers[0].Output.SizeY;
    NumberOfClasses := NN.GetLastLayer().Output.Size;

    NeuralFit := TNeuralImageFit.Create;
    vInputImage := TNNetVolume.Create();
    vOutput := TNNetVolume.Create(NumberOfClasses);

    ImageFileName := 'plant/Apple___Black_rot/image (1).JPG';
    WriteLn('Loading image: ',ImageFileName);

    if LoadImageFromFileIntoVolume(
      ImageFileName, vInputImage, InputSizeX, InputSizeY,
      {EncodeNeuronalInput=}csEncodeRGB) then
    begin
      WriteLn('Classifying the image:', ImageFileName);
      vOutput.Fill(0);
      NeuralFit.ClassifyImage(NN, vInputImage, vOutput);
      WriteLn('The image belongs to the class of images: ', vOutput.GetClass());
    end
    else
    begin
      WriteLn('Failed loading image: ',ImageFileName);
    end;

    vInputImage.Free;
    vOutput.Free;
    NeuralFit.Free;

    NN.Free;
  end;

The trained neural network (model) is loaded with

    NN := TNNet.Create;
    NN.LoadFromFile('SimplePlantLeafDisease-20230720.nn');

The input image size is found from the loaded model with:

    InputSizeX := NN.Layers[0].Output.SizeX;
    InputSizeY := NN.Layers[0].Output.SizeY;

The number of classes is found from the loaded model with:

    NumberOfClasses := NN.GetLastLayer().Output.Size;

The image is loaded, resized and scaled from [0,255] to [-2,+2] with:

    ImageFileName := 'plant/Apple___Black_rot/image (1).JPG';
    WriteLn('Loading image: ',ImageFileName);

    if LoadImageFromFileIntoVolume(
      ImageFileName, vInputImage, InputSizeX, InputSizeY,
      {EncodeNeuronalInput=}csEncodeRGB) then       

The NN is run with plenty of tricks specific for computer vision with:

      NeuralFit.ClassifyImage(NN, vInputImage, vOutput);

The output of the neural network is placed at vOutput. The actual predicted class can be found with:

      vOutput.GetClass()