org.datavec.image.transform.ImageTransform Java Examples

@Override
public RecordReader getRecordReader(long rngSeed, int[] imgDim, DataSetType set, ImageTransform imageTransform) {
    try {
        Random rng = new Random(rngSeed);
        File datasetPath = getDataSetPath(set);

        FileSplit data = new FileSplit(datasetPath, BaseImageLoader.ALLOWED_FORMATS, rng);
        ObjectDetectionRecordReader recordReader = new ObjectDetectionRecordReader(imgDim[1], imgDim[0], imgDim[2],
                        imgDim[4], imgDim[3], null);

        recordReader.initialize(data);
        return recordReader;
    } catch (IOException e) {
        throw new RuntimeException("Could not download SVHN", e);
    }
}
 

/**
 * Get the image loader
 * configuring it using the
 * {@link ConverterArgs}
 *
 * @param input         the input to convert
 * @param converterArgs the converter args to use
 * @return the configured {@link NativeImageLoader}
 */
public NativeImageLoader getImageLoader(T input, ConverterArgs converterArgs) {
    if (converterArgs == null || converterArgs.getLongs().isEmpty())
        return new NativeImageLoader();
    else if (converterArgs.getLongs().size() == 3) {
        if (converterArgs.getImageTransformProcess() != null) {
            return new NativeImageLoader(converterArgs.getLongs().get(0), converterArgs.getLongs().get(1), converterArgs.getLongs().get(2), new MultiImageTransform(converterArgs.getImageTransformProcess().getTransformList().toArray(new ImageTransform[1])));

        } else if (converterArgs.getImageTransformProcess() != null)
            return new NativeImageLoader(converterArgs.getLongs().get(0), converterArgs.getLongs().get(1), converterArgs.getLongs().get(2), new MultiImageTransform(converterArgs.getImageTransformProcess().getTransformList().toArray(new ImageTransform[1])));
        else {
            return new NativeImageLoader(converterArgs.getLongs().get(0), converterArgs.getLongs().get(1), converterArgs.getLongs().get(2));

        }
    } else if (converterArgs.getLongs().size() == 3) {
        if (converterArgs.getImageTransformProcess() != null) {
            return new NativeImageLoader(converterArgs.getLongs().get(0).intValue(), converterArgs.getLongs().get(1).intValue()
                    , converterArgs.getLongs().get(2).intValue(), new MultiImageTransform(converterArgs.getImageTransformProcess().getTransformList().toArray(new ImageTransform[1])));
        } else
            return new NativeImageLoader(converterArgs.getLongs().get(0).intValue(), converterArgs.getLongs().get(1).intValue()
                    , converterArgs.getLongs().get(2).intValue());
    }


    return new NativeImageLoader();
}
 

/**
 * Preprocess and store cifar based on successful Torch approach by Sergey Zagoruyko
 * Reference: https://github.com/szagoruyko/cifar.torch
 */
public opencv_core.Mat convertCifar(Mat orgImage) {
    numExamples++;
    Mat resImage = new Mat();
    OpenCVFrameConverter.ToMat converter = new OpenCVFrameConverter.ToMat();
    //        ImageTransform yuvTransform = new ColorConversionTransform(new Random(seed), COLOR_BGR2Luv);
    //        ImageTransform histEqualization = new EqualizeHistTransform(new Random(seed), COLOR_BGR2Luv);
    ImageTransform yuvTransform = new ColorConversionTransform(new Random(seed), COLOR_BGR2YCrCb);
    ImageTransform histEqualization = new EqualizeHistTransform(new Random(seed), COLOR_BGR2YCrCb);

    if (converter != null) {
        ImageWritable writable = new ImageWritable(converter.convert(orgImage));
        // TODO determine if need to normalize y before transform - opencv docs rec but currently doing after
        writable = yuvTransform.transform(writable); // Converts to chrome color to help emphasize image objects
        writable = histEqualization.transform(writable); // Normalizes values to further clarify object of interest
        resImage = converter.convert(writable.getFrame());
    }

    return resImage;
}
 

/**
 * Preprocess and store cifar based on successful Torch approach by Sergey Zagoruyko
 * Reference: <a href="https://github.com/szagoruyko/cifar.torch">https://github.com/szagoruyko/cifar.torch</a>
 */
public Mat convertCifar(Mat orgImage) {
    numExamples++;
    Mat resImage = new Mat();
    OpenCVFrameConverter.ToMat converter = new OpenCVFrameConverter.ToMat();
    //        ImageTransform yuvTransform = new ColorConversionTransform(new Random(seed), COLOR_BGR2Luv);
    //        ImageTransform histEqualization = new EqualizeHistTransform(new Random(seed), COLOR_BGR2Luv);
    ImageTransform yuvTransform = new ColorConversionTransform(new Random(seed), COLOR_BGR2YCrCb);
    ImageTransform histEqualization = new EqualizeHistTransform(new Random(seed), COLOR_BGR2YCrCb);

    if (converter != null) {
        ImageWritable writable = new ImageWritable(converter.convert(orgImage));
        // TODO determine if need to normalize y before transform - opencv docs rec but currently doing after
        writable = yuvTransform.transform(writable); // Converts to chrome color to help emphasize image objects
        writable = histEqualization.transform(writable); // Normalizes values to further clarify object of interest
        resImage = converter.convert(writable.getFrame());
    }

    return resImage;
}
 

public CaptchaLoader(
    int height, int width, int channels, ImageTransform imageTransform, String dataSetType) {
  super(height, width, channels, imageTransform);
  this.height = height;
  this.width = width;
  this.channels = channels;
  try {
    this.fullDir = new File("src/main/resources");
    logger.info("fullDir: " + fullDir);
  } catch (Exception e) {
    logger.error("The datasets directory failed, please checking.", e);
    throw new RuntimeException(e);
  }
  this.fullDir = new File(fullDir, dataSetType);
  load();
}
 

public CifarLoader(int height, int width, int channels, ImageTransform imgTransform, boolean train,
                boolean useSpecialPreProcessCifar, File fullDir, long seed, boolean shuffle) {
    super(height, width, channels, imgTransform);
    this.height = height;
    this.width = width;
    this.channels = channels;
    this.train = train;
    this.useSpecialPreProcessCifar = useSpecialPreProcessCifar;
    this.seed = seed;
    this.shuffle = shuffle;

    if (fullDir == null) {
        this.fullDir = getDefaultDirectory();
    } else {
        this.fullDir = fullDir;
    }
    meanVarPath = new File(this.fullDir, "meanVarPath.txt");
    trainFilesSerialized = FilenameUtils.concat(this.fullDir.toString(), "cifar_train_serialized");
    testFilesSerialized = FilenameUtils.concat(this.fullDir.toString(), "cifar_test_serialized.ser");

    load();
}
 

public CifarLoader(int height, int width, int channels, ImageTransform imgTransform, boolean train,
                boolean useSpecialPreProcessCifar, File fullDir, long seed, boolean shuffle) {
    super(height, width, channels, imgTransform);
    this.height = height;
    this.width = width;
    this.channels = channels;
    this.train = train;
    this.useSpecialPreProcessCifar = useSpecialPreProcessCifar;
    this.seed = seed;
    this.shuffle = shuffle;

    if (fullDir == null) {
        this.fullDir = getDefaultDirectory();
    } else {
        this.fullDir = fullDir;
    }
    meanVarPath = new File(this.fullDir, "meanVarPath.txt");
    trainFilesSerialized = FilenameUtils.concat(this.fullDir.toString(), "cifar_train_serialized");
    testFilesSerialized = FilenameUtils.concat(this.fullDir.toString(), "cifar_test_serialized.ser");

    load();
}
 

protected BaseImageRecordReader(int height, int width, int channels, PathLabelGenerator labelGenerator,
                                PathMultiLabelGenerator labelMultiGenerator, ImageTransform imageTransform) {
    this.height = height;
    this.width = width;
    this.channels = channels;
    this.labelGenerator = labelGenerator;
    this.labelMultiGenerator = labelMultiGenerator;
    this.imageTransform = imageTransform;
    this.appendLabel = (labelGenerator != null || labelMultiGenerator != null);
}
 

public LFWLoader(int[] imgDim, ImageTransform imgTransform, boolean useSubset) {
    this.height = imgDim[0];
    this.width = imgDim[1];
    this.channels = imgDim[2];
    this.imageTransform = imgTransform;
    this.useSubset = useSubset;
    this.localDir = useSubset ? localSubDir : localDir;
    this.fullDir = new File(BASE_DIR, localDir);
    generateLfwMaps();
}
 

protected BaseImageRecordReader(long height, long width, long channels, boolean nchw_channels_first, PathLabelGenerator labelGenerator,
                                PathMultiLabelGenerator labelMultiGenerator, ImageTransform imageTransform) {
    this.height = height;
    this.width = width;
    this.channels = channels;
    this.labelGenerator = labelGenerator;
    this.labelMultiGenerator = labelMultiGenerator;
    this.imageTransform = imageTransform;
    this.appendLabel = (labelGenerator != null || labelMultiGenerator != null);
    this.nchw_channels_first = nchw_channels_first;
}
 

private static ImageTransform[] getTransforms() {
	ImageTransform randCrop = new CropImageTransform(new Random(), 10);
	ImageTransform warpTransform = new WarpImageTransform(new Random(), 42);
	ImageTransform flip = new FlipImageTransform(new Random());
	ImageTransform scale = new ScaleImageTransform(new Random(), 1);
	return new ImageTransform[] { randCrop, warpTransform, flip, scale };
}
 

public LFWLoader(int[] imgDim, ImageTransform imgTransform, boolean useSubset) {
    this.height = imgDim[0];
    this.width = imgDim[1];
    this.channels = imgDim[2];
    this.imageTransform = imgTransform;
    this.useSubset = useSubset;
    this.localDir = useSubset ? localSubDir : localDir;
    this.fullDir = new File(BASE_DIR, localDir);
    generateLfwMaps();
}
 

public LFWLoader(long[] imgDim, ImageTransform imgTransform, boolean useSubset) {
    this.height = imgDim[0];
    this.width = imgDim[1];
    this.channels = imgDim[2];
    this.imageTransform = imgTransform;
    this.useSubset = useSubset;
    this.localDir = useSubset ? localSubDir : localDir;
    this.fullDir = new File(BASE_DIR, localDir);
    generateLfwMaps();
}
 

@Override
public CSVSequenceRecordReader getRecordReader(long rngSeed, int[] shape, DataSetType set, ImageTransform transform) {
    return getRecordReader(rngSeed, set);
}
 

public Java2DNativeImageLoader(int height, int width, int channels, ImageTransform imageTransform) {
    super(height, width, channels, imageTransform);
}
 

public AndroidNativeImageLoader(int height, int width, int channels, ImageTransform imageTransform) {
    super(height, width, channels, imageTransform);
}
 

public CifarLoader(int height, int width, int channels, ImageTransform imgTransform, boolean train,
                boolean useSpecialPreProcessCifar) {
    this(height, width, channels, imgTransform, train, useSpecialPreProcessCifar, DEFAULT_SHUFFLE);
}
 

public CifarLoader(int height, int width, int channels, ImageTransform imgTransform, boolean train,
                boolean useSpecialPreProcessCifar, boolean shuffle) {
    this(height, width, channels, imgTransform, train, useSpecialPreProcessCifar, null, System.currentTimeMillis(),
                    shuffle);
}
 

protected BaseImageRecordReader(long height, long width, long channels, PathLabelGenerator labelGenerator,
                                PathMultiLabelGenerator labelMultiGenerator, ImageTransform imageTransform) {
    this(height, width, channels, true, labelGenerator, labelMultiGenerator, imageTransform);
}
 

public BaseImageRecordReader(long height, long width, long channels, PathLabelGenerator labelGenerator,
                             ImageTransform imageTransform) {
    this(height, width, channels, labelGenerator, null, imageTransform);
}
 

/**
 * As per {@link #ObjectDetectionRecordReader(int, int, int, int, int, boolean, ImageObjectLabelProvider, ImageTransform)}
 * but hardcoded to NCHW format
 */
public ObjectDetectionRecordReader(int height, int width, int channels, int gridH, int gridW,
                                   ImageObjectLabelProvider labelProvider, ImageTransform imageTransform) {
    this(height, width, channels, gridH, gridW, true, labelProvider, imageTransform);
}
 

/** Loads images with given height, width, and channels, appending no labels.
 * Output format is NCHW (channels first) - [numExamples, channels, height, width]*/
public ImageRecordReader(long height, long width, long channels, ImageTransform imageTransform) {
    super(height, width, channels, null, imageTransform);
}
 

/** Loads images with given height, width, and channels, appending labels returned by the generator.<br>
 * If {@code nchw_channels_first == true} output format is NCHW (channels first) - [numExamples, channels, height, width]<br>
 * If {@code nchw_channels_first == false} output format is NHWC (channels last) - [numExamples, height, width, channels]<br>
 */
public ImageRecordReader(long height, long width, long channels, boolean nchw_channels_first, PathLabelGenerator labelGenerator,
                         ImageTransform imageTransform) {
    super(height, width, channels, nchw_channels_first, labelGenerator, null, imageTransform);
}
 

public LegacyImageTransformDeserializer() {
    super(ImageTransform.class, LegacyMappingHelper.getLegacyMappingImageTransform());
}
 

public Java2DNativeImageLoader(int height, int width, int channels, ImageTransform imageTransform) {
    super(height, width, channels, imageTransform);
}
 

public CifarLoader(int height, int width, int channels, ImageTransform imgTransform, boolean train,
                boolean useSpecialPreProcessCifar, boolean shuffle) {
    this(height, width, channels, imgTransform, train, useSpecialPreProcessCifar, null, System.currentTimeMillis(),
                    shuffle);
}
 

public Java2DNativeImageLoader(int height, int width, int channels, ImageTransform imageTransform) {
    super(height, width, channels, imageTransform);
}
 

public CifarLoader(int height, int width, int channels, ImageTransform imgTransform, boolean train,
                boolean useSpecialPreProcessCifar) {
    this(height, width, channels, imgTransform, train, useSpecialPreProcessCifar, DEFAULT_SHUFFLE);
}
 

public CaptchaSetIterator(int batchSize, ImageTransform imageTransform, String dataSetType) {
  this.batchSize = batchSize;
  load = new CaptchaLoader(imageTransform, dataSetType);
  numExample = load.totalExamples();
}
 

public AndroidNativeImageLoader(int height, int width, int channels, ImageTransform imageTransform) {
    super(height, width, channels, imageTransform);
}