paddlers/docs/intro/transforms_cons_params_en.md

简体中文 | English

PaddleRS Data Transformation Operator Construction Parameters

This document describes the parameters of each PaddleRS data transformation operator, including the operator name, operator purpose, parameter name, parameter type, parameter meaning, and parameter default value of each operator.

You can check all data transformation operators supported by PaddleRS here.

AppendIndex

Append remote sensing index to input image(s).

Parameter Name (Parameter Type)	Description	Default Value
index_type (str)	Type of remote sensinng index. See supported index types in https://github.com/PaddlePaddle/PaddleRS/tree/develop/paddlers/transforms/indices.py
band_indexes (dict)	Mapping of band names to band indices (starting from 1). See supported band names in https://github.com/PaddlePaddle/PaddleRS/tree/develop/paddlers/transforms/indices.py	None
satellite (str)	Type of satellite. If set, band indices will be automatically determined accordingly. See supported satellites in https://github.com/PaddlePaddle/PaddleRS/tree/develop/paddlers/transforms/satellites.py	None

CenterCrop

Crop the input image(s) at the center.

1. Locate the center of the image.
2. Crop the image.

Parameter Name (Parameter Type)	Description	Default Value
crop_size (int)	Target size of the cropped image(s)	224

Dehaze

Dehaze input image(s)

Parameter Name (Parameter Type)	Description	Default Value
gamma (bool)	Use gamma correction or not	False

MatchRadiance

Perform relative radiometric correction between bi-temporal images.

Parameter Name (Parameter Type)	Description	Default Value
method (str)	Method used to match the radiance of the bi-temporal images. Choices are {'hist', 'lsr', 'fft'}. 'hist' stands for histogram matching, 'lsr' stands for least-squares regression, and 'fft' replaces the low-frequency components of the image to match the reference image	'hist'

MixupImage

Mixup two images and their gt_bbox/gt_score.

Parameter Name (Parameter Type)	Description	Default Value
alpha (float)	Alpha parameter of beta distribution	1.5
beta (float)	Beta parameter of beta distribution	1.5

Normalize

Apply normalization to the input image(s). The normalization steps are:

1. im = (im - min_value) * 1 / (max_value - min_value)
2. im = im - mean
3. im = im / std

Parameter Name (Parameter Type)	Description	Default Value
mean (list[float] \| tuple[float])	Mean of input image(s)	[0.485,0.456,0.406]
std (list[float] \| tuple[float])	Standard deviation of input image(s)	[0.229,0.224,0.225]
min_val (list[float] \| tuple[float])	Inimum value of input image(s). If None, use 0 for all channels	None
max_val (list[float] \| tuple[float])	Maximum value of input image(s). If None, use 255 for all channels	None
apply_to_tar (bool) | Whether to apply transformation to the target image	True

Pad

Pad image to a specified size or multiple of size_divisor.

Parameter Name (Parameter Type)	Description	Default Value
target_size (list[int] \| tuple[int])	Image target size, if None, pad to multiple of size_divisor	None
pad_mode (int)	Currently only four modes are supported:[-1, 0, 1, 2]. if -1, use specified offsets. If 0, only pad to right and bottom. If 1, pad according to center. If 2, only pad left and top	0
offset (list[int] \| None)	Padding offsets	None
im_padding_value (list[float] \| tuple[float])	RGB value of padded area	(127.5,127.5,127.5)
label_padding_value (int)	Filling value for the mask	255
size_divisor (int)	Image width and height after padding will be a multiple of size_divisor

RandomBlur

Randomly blur input image(s).

Parameter Name (Parameter Type)	Description	Default Value
prob (float)	Probability of blurring

RandomCrop

Randomly crop the input.

1. Compute the height and width of cropped area according to aspect_ratio andscaling.
2. Locate the upper left corner of cropped area randomly.
3. Crop the image(s).
4. Resize the cropped area to crop_size x crop_size.

Parameter Name (Parameter Type)	Description	Default Value
crop_size (int \| list[int] \| tuple[int])	Target size of the cropped area. If None, the cropped area will not be resized	None
aspect_ratio (list[float])	Aspect ratio of cropped region in [min, max] format	[.5, 2.]
thresholds (list[float])	IoU thresholds to decide a valid bbox crop	[.0,.1, .3, .5, .7, .9]
scaling (list[float])	Ratio between the cropped region and the original image in [min, max] format	[.3, 1.]
num_attempts (int)	Max number of tries before giving up	50
allow_no_crop (bool)	Whether returning without doing crop is allowed	True
cover_all_box (bool)	Whether to force to cover the entire target box	False

RandomDistort

Random color distortion.

Parameter Name (Parameter Type)	Description	Default Value
brightness_range (float)	Range of brightness distortion	.5
brightness_prob (float)	Probability of brightness distortion	.5
contrast_range (float)	Range of contrast distortion	.5
contrast_prob (float)	Probability of contrast distortion	.5
saturation_range (float,optional)	Range of saturation distortion	.5
saturation_prob (float)	Probability of saturation distortion	.5
hue_range (float)	Range of hue distortion	.5
hue_prob (float)	Probability of hue distortion	.5
random_apply (bool)	Apply the transformation in random (YOLO) or fixed (SSD) order	True
count (int)	Count used to control the distortion	4
shuffle_channel (bool)	Whether to permute channels randomly	False

RandomExpand

Randomly expand the input by padding according to random offsets.

Parameter Name (Parameter Type)	Description	Default Value
upper_ratio (float)	Maximum ratio to which the original image is expanded	4
prob (float)	Probability of expanding	.5
im_padding_value (list[float] \| tuple[float])	RGB filling value for the image	(127.5,127.5,127.5)
label_padding_value (int)	Filling value for the mask	255

RandomHorizontalFlip

Randomly flip the input horizontally.

Parameter Name (Parameter Type)	Description	Default Value
prob (float)	Probability of flipping the input	.5

RandomResize

Resize input to random sizes.

Attention: If interp is 'RANDOM', the interpolation method will be chosen randomly.

Parameter Name (Parameter Type)	Description	Default Value
Target_sizes (list[int] \| list[list \| tuple] \| tuple [list \| tuple])	Multiple target sizes, each of which should be int, list, or tuple	.5
interp (str)	Interpolation method for resizing image(s). One of {'NEAREST', 'LINEAR', 'CUBIC', 'AREA', 'LANCZOS4', 'RANDOM'}	'LINEAR'

RandomResizeByShort

Resize input to random sizes while keeping the aspect ratio.

Attention: If interp is 'RANDOM', the interpolation method will be chosen randomly.

Parameter Name (Parameter Type)	Description	Default Value
short_sizes (int \| list[int])	Target size of the shorter side of the image(s)	.5
max_size (int)	Upper bound of longer side of the image(s). If max_size is -1, no upper bound will be applied	-1
interp (str)	Interpolation method for resizing image(s). One of {'NEAREST', 'LINEAR', 'CUBIC', 'AREA', 'LANCZOS4', 'RANDOM'}	'LINEAR'

RandomScaleAspect

Crop input image(s) and resize back to original sizes.

Parameter Name (Parameter Type)	Description	Default Value
min_scale (float)	Minimum ratio between the cropped region and the original image. If 0, image(s) will not be cropped	0
aspect_ratio (float)	Aspect ratio of cropped region	.33

RandomSwap

Randomly swap multi-temporal images.

Parameter Name (Parameter Type)	Description	Default Value
prob (float)	Probability of swapping the input images	0.2

RandomVerticalFlip

Randomly flip the input vertically.

Parameter Name (Parameter Type)	Description	Default Value
prob (float)	Probability of flipping the input	.5

ReduceDim

Use PCA to reduce the dimension of input image(s).

Parameter Name (Parameter Type)	Description	Default Value
joblib_path (str)	Path of *.joblib file of PCA
apply_to_tar (bool)	Whether to apply transformation to the target image	True

Resize

Resize input.

• If target_size is an int, resize the image(s) to target_size x target_size.
• If target_size is a list or tuple, resize the image(s) to target_size.

Attention: If interp is 'RANDOM', the interpolation method will be chosen randomly.

Parameter Name (Parameter Type)	Description	Default Value
target_size (int \| list[int] \| tuple[int])	Target size. If it is an integer, the target height and width will be both set to target_size. Otherwise, target_size represents [target height, target width]
interp (str)	Interpolation method for resizing image(s). One of {'NEAREST', 'LINEAR', 'CUBIC', 'AREA', 'LANCZOS4', 'RANDOM'}	'LINEAR'
keep_ratio (bool)	If True, the scaling factor of width and height will be set to same value, and height/width of the resized image will be no greater than the target width/height	False

ResizeByLong

Resize the input image, keeping the aspect ratio unchanged (calculate the scaling factor based on the long side).

Attention: If interp is 'RANDOM', the interpolation method will be chosen randomly.

Parameter Name (Parameter Type)	Description	Default Value
long_size (int)	The size of the target on the longer side of the image
interp (str)	Interpolation method for resizing image(s). One of {'NEAREST', 'LINEAR', 'CUBIC', 'AREA', 'LANCZOS4', 'RANDOM'}	'LINEAR'

ResizeByShort

Resize input while keeping the aspect ratio.

Attention: If interp is 'RANDOM', the interpolation method will be chosen randomly.

Parameter Name (Parameter Type)	Description	Default Value
short_size (int)	Target size of the shorter side of the image(s)
max_size (int)	Upper bound of longer side of the image(s). If max_size is -1, no upper bound will be applied	-1
interp (str)	Interpolation method for resizing image(s). One of {'NEAREST', 'LINEAR', 'CUBIC', 'AREA', 'LANCZOS4', 'RANDOM'}	'LINEAR'

SelectBand

Select a set of bands of input image(s).

Parameter Name (Parameter Type)	Description	Default Value
band_list (list)	Bands to select (band index starts from 1)	[1, 2, 3]
apply_to_tar (bool)	Whether to apply transformation to the target image	True