Python v0.3.0.dev+adb6fa6
  • Overview

    • CropMirrorNormalize

    class dragon.vm.dali.ops.CropMirrorNormalize[source]

    Crop and normalize input with the horizontal flip.

    Examples:

    flip_rng = dali.ops.CoinFlip(0.5)
cmn = dali.ops.CropMirrorNormalize(
    # Match the number of spatial dims
    # (H, W) for 2d input
    # (D, H, W) for 3d input
    crop=(224, 224),
    # Historical values to normalize input
    mean=(102., 115., 122.),
    std=(1., 1., 1.),
    # ``BGR``, ``RGB``, or ``GRAY``
    image_type='BGR',
    # Or ``float16`` for fp16 training
    output_dtype='float32',
    # Or ``NHWC``
    output_layout='NCHW'
)
y = cmn(inputs['x'], mirror=flip_rng())

    __new__

    static CropMirrorNormalize.__new__(
      cls,
      crop=None,
      mirror=None,
      mean=0.0,
      std=1.0,
      image_type='BGR',
      output_dtype='float32',
      output_layout='NCHW'
    )[source]

    Create a CropMirrorNormalize operator.

    Parameters:
    • crop (Sequence[int], optional) – The cropped spatial dimensions for output.
    • mirror ({0, 1}, optional) – Whether to apply the horizontal flip.
    • mean (Union[float, Sequence[float]], optional) – The values to subtract.
    • std (Union[float, Sequence[float]], optional) – The values to divide after subtraction.
    • image_type ({'BGR', 'RGB'}, optional) – The color space of input.
    • output_dtype ({'float16', 'float32'}, optional) – The data type of output.
    • output_layout ({'NCHW', 'NHWC'}, optional) – The data format of output.
    Returns:

    nvidia.dali.ops.CropMirrorNormalize – The operator.