Package smile.vision.transform

Interface Transform

public interface Transform
Transformation from image to tensor.

  • Field Summary

    Fields
    Modifier and Type
    Field
    Description
    static final float[]
    DEFAULT_MEAN
    The default mean value of pixel RGB after normalized to [0, 1].
    static final float[]
    DEFAULT_STD
    The default standard deviation of pixel RGB after normalized to [0, 1].

  • Method Summary

    Modifier and Type
    Method
    Description
    static Transform
    classification(int cropSize, int resizeSize)
    Returns a transform for image classification.
    static Transform
    classification(int cropSize, int resizeSize, float[] mean, float[] std, int hints)
    Returns a transform for image classification.
    default BufferedImage
    crop(BufferedImage image, int size, boolean deep)
    Crops an image.
    default BufferedImage
    crop(BufferedImage image, int width, int height, boolean deep)
    Crops an image.
    Tensor
    forward(BufferedImage... images)
    Transforms images to 4-D tensor with shape [samples, channels, height, width].
    default BufferedImage
    resize(BufferedImage image, int size, int hints)
    Resizes an image and keeps the aspect ratio.
    default Tensor
    toTensor(float[] mean, float[] std, BufferedImage... images)
    Returns the tensor with NCHW shape [samples, channels, height, width] of the images.

  • Field Details

    • DEFAULT_MEAN

      static final float[] DEFAULT_MEAN
      The default mean value of pixel RGB after normalized to [0, 1]. Calculated on ImageNet data.

    • DEFAULT_STD

      static final float[] DEFAULT_STD
      The default standard deviation of pixel RGB after normalized to [0, 1]. Calculated on ImageNet data.

  • Method Details

    • forward

      Tensor forward(BufferedImage... images)
      Transforms images to 4-D tensor with shape [samples, channels, height, width].
      Parameters:
      images - the input images.
      Returns:
      the 4-D tensor representation of images.

    • resize

      default BufferedImage resize(BufferedImage image, int size, int hints)
      Resizes an image and keeps the aspect ratio.
      Parameters:
      image - the input image.
      size - the image size of the shorter side.
      hints - flags to indicate the type of algorithm to use for image resampling. See SCALE_DEFAULT, SCALE_FAST, SCALE_SMOOTH, SCALE_REPLICATE, SCALE_AREA_AVERAGING of java.awt.Image class.
      Returns:
      the output image.

    • crop

      default BufferedImage crop(BufferedImage image, int size, boolean deep)
      Crops an image.
      Parameters:
      image - the image size.
      size - the cropped image size.
      deep - If false, the returned BufferedImage shares the same data array as the original image. Otherwise, returns a deep copy.
      Returns:
      the cropped image.

    • crop

      default BufferedImage crop(BufferedImage image, int width, int height, boolean deep)
      Crops an image. The returned BufferedImage shares the same data array as the original image.
      Parameters:
      image - the image size.
      width - the cropped image width.
      height - the cropped image height.
      deep - If false, the returned BufferedImage shares the same data array as the original image. Otherwise, returns a deep copy.
      Returns:
      the cropped image.

    • toTensor

      default Tensor toTensor(float[] mean, float[] std, BufferedImage... images)
      Returns the tensor with NCHW shape [samples, channels, height, width] of the images. The values of tensor are first rescaled to [0.0, 1.0] and then normalized.
      Parameters:
      mean - the normalization mean.
      std - the normalization standard deviation.
      images - the input images that should have same size.
      Returns:
      the output tensor.

    • classification

      static Transform classification(int cropSize, int resizeSize)
      Returns a transform for image classification. The images are resized using scaling hints, followed by a central crop. Finally, the values are first rescaled to [0.0, 1.0] and then normalized.
      Parameters:
      cropSize - the crop size.
      resizeSize - the scaling size.
      Returns:
      a transform for image classification.

    • classification

      static Transform classification(int cropSize, int resizeSize, float[] mean, float[] std, int hints)
      Returns a transform for image classification. The images are resized using scaling hints, followed by a central crop. Finally, the values are first rescaled to [0.0, 1.0] and then normalized.
      Parameters:
      cropSize - the crop size.
      resizeSize - the scaling size.
      mean - the normalization mean.
      std - the normalization standard deviation.
      hints - the scaling hints.
      Returns:
      a transform for image classification.