neural_compressor.data.transforms.imagenet_transform

Neural Compressor built-in imagenet transforms.

Module Contents

Classes

QuantizedInput

Convert the dtype of input to quantize it.

LabelShift

Convert label to label - label_shift.

ParseDecodeImagenet

Parse features in Example proto.

ParseDecodeImagenetTransform

Imagenet decoding will be performed automatically from Neural Compressor v1.4.

TensorflowResizeCropImagenetTransform

Combination of a series of transforms which is applicable to images in Imagenet.

BilinearImagenetTransform

Combination of a series of transforms which is applicable to images in Imagenet.

OnnxBilinearImagenetTransform

Combination of a series of transforms which is applicable to images in Imagenet.

ONNXResizeCropImagenetTransform

Combination of a series of transforms which is applicable to images in Imagenet.

ResizeWithAspectRatio

Resize the image with aspect ratio.
class neural_compressor.data.transforms.imagenet_transform.QuantizedInput(dtype, scale=None)

Bases: neural_compressor.data.transforms.transform.BaseTransform

Convert the dtype of input to quantize it.

Parameters:

  • dtype (str) – desired image dtype, support ‘uint8’, ‘int8’

  • scale (float, default=None) – scaling ratio of each point in image

Returns:

tuple of processed image and label

class neural_compressor.data.transforms.imagenet_transform.LabelShift(label_shift=0)

Bases: neural_compressor.data.transforms.transform.BaseTransform

Convert label to label - label_shift.

Parameters:

label_shift (int, default=0) – number of label shift

Returns:

tuple of processed image and label

class neural_compressor.data.transforms.imagenet_transform.ParseDecodeImagenet

Parse features in Example proto.

Returns:

tuple of parsed image and label

class neural_compressor.data.transforms.imagenet_transform.ParseDecodeImagenetTransform

Bases: neural_compressor.data.transforms.transform.BaseTransform

Imagenet decoding will be performed automatically from Neural Compressor v1.4.

Returns:

sample

class neural_compressor.data.transforms.imagenet_transform.TensorflowResizeCropImagenetTransform(height, width, random_crop=False, resize_side=256, resize_method='bilinear', random_flip_left_right=False, mean_value=[0.0, 0.0, 0.0], scale=1.0, data_format='channels_last', subpixels='RGB')

Bases: neural_compressor.data.transforms.transform.BaseTransform

Combination of a series of transforms which is applicable to images in Imagenet.

Parameters:

  • height (int) – Height of the result

  • width (int) – Width of the result

  • random_crop (bool, default=False) – whether to random crop

  • resize_side (int, default=256) – desired shape after resize operation

  • random_flip_left_right (bool, default=False) – whether to random flip left and right

  • mean_value (list, default=[0.0,0.0,0.0]) – means for each channel

  • scale (float, default=1.0) – std value

Returns:

tuple of processed image and label

class neural_compressor.data.transforms.imagenet_transform.BilinearImagenetTransform(height, width, central_fraction=0.875, mean_value=[0.0, 0.0, 0.0], scale=1.0)

Bases: neural_compressor.data.transforms.transform.BaseTransform

Combination of a series of transforms which is applicable to images in Imagenet.

Parameters:

  • height – Height of the result

  • width – Width of the result

  • central_fraction (float, default=0.875) – fraction of size to crop

  • mean_value (list, default=[0.0,0.0,0.0]) – means for each channel

  • scale (float, default=1.0) – std value

Returns:

tuple of processed image and label

class neural_compressor.data.transforms.imagenet_transform.OnnxBilinearImagenetTransform(height, width, central_fraction=0.875, mean_value=[0.0, 0.0, 0.0], scale=1.0)

Bases: neural_compressor.data.transforms.transform.BaseTransform

Combination of a series of transforms which is applicable to images in Imagenet.

Parameters:

  • height – Height of the result

  • width – Width of the result

  • central_fraction (float, default=0.875) – fraction of size to crop

  • mean_value (list, default=[0.0,0.0,0.0]) – means for each channel

  • scale (float, default=1.0) – std value

Returns:

tuple of processed image and label

class neural_compressor.data.transforms.imagenet_transform.ONNXResizeCropImagenetTransform(height, width, random_crop=False, resize_side=256, mean_value=[0.0, 0.0, 0.0], std_value=[0.229, 0.224, 0.225], resize_method='bilinear', data_format='channels_last', subpixels='RGB')

Bases: neural_compressor.data.transforms.transform.BaseTransform

Combination of a series of transforms which is applicable to images in Imagenet.

Parameters:

  • height – Height of the result

  • width – Width of the result

  • central_fraction (float, default=0.875) – fraction of size to crop

  • mean_value (list, default=[0.0,0.0,0.0]) – means for each channel

  • scale (float, default=1.0) – std value

Returns:

tuple of processed image and label

class neural_compressor.data.transforms.imagenet_transform.ResizeWithAspectRatio(height, width, scale=87.5, inter_pol=cv2.INTER_AREA)

Bases: neural_compressor.data.transforms.transform.BaseTransform

Resize the image with aspect ratio.

Returns:

image and label