Geometric
Processors for geometric image transformations
RandomFlipBoxesLeftRight
paz.processors.geometric.RandomFlipBoxesLeftRight()
Flips image and implemented labels horizontally.
ToImageBoxCoordinates
paz.processors.geometric.ToImageBoxCoordinates()
Convert normalized box coordinates to image-size box coordinates.
ToNormalizedBoxCoordinates
paz.processors.geometric.ToNormalizedBoxCoordinates()
Convert image-size box coordinates to normalized box coordinates.
RandomSampleCrop
paz.processors.geometric.RandomSampleCrop(probability=0.5, max_trials=50)
Crops image while adjusting the normalized corner form bounding boxes.
Arguments
- probability: Float between ''[0, 1]''.
RandomTranslation
paz.processors.geometric.RandomTranslation(delta_scale=[0.25, 0.25], fill_color=None)
Applies a random translation to image and labels
Arguments
- delta_scale: List with two elements having the normalized deltas. e.g. ''[.25, .25]''.
fill_color: List of three integers indicating the color values e.g. ''[0, 0, 0]''.
RandomRotation
paz.processors.geometric.RandomRotation(rotation_range=30, fill_color=None, probability=0.5)
Randomly rotate an images
Arguments
- rotation_range: Int. indicating the max and min values in degrees
of the uniform distribution
[-range, range]
from which the angles are sampled. - fill_color: ''None'' or List of three integers indicating the
color values e.g.
[0, 0, 0]
. IfNone
mean channel values of the image will be calculated as fill values. - probability: Float between 0 and 1.
RandomKeypointTranslation
paz.processors.geometric.RandomKeypointTranslation(delta_scale=[0.2, 0.2], fill_color=None, probability=0.5)
Applies a random translation to image and keypoints.
Arguments
- delta_scale: List with two elements having the normalized deltas. e.g. ''[.25, .25]''.
- fill_color: ''None'' or List of three integers indicating the color values e.g. ''[0, 0, 0]''. If ''None'' mean channel values of the image will be calculated as fill values.
- probability: Float between ''[0, 1]''.
RandomKeypointRotation
paz.processors.geometric.RandomKeypointRotation(rotation_range=30, fill_color=None, probability=0.5)
Randomly rotate an images with its corresponding keypoints.
Arguments
- rotation_range: Int. indicating the max and min values in degrees of the uniform distribution ''[-range, range]'' from which the angles are sampled.
- fill_color: ''None'' or List of three integers indicating the color values e.g. ''[0, 0, 0]''. If ''None'' mean channel values of the image will be calculated as fill values.
GetTransformationSize
paz.processors.geometric.GetTransformationSize(input_size, multiple)
Calculate the transformation size for the imgae. The size is tuple of length two indicating the x, y values.
Arguments
- image: Numpy array
GetTransformationScale
paz.processors.geometric.GetTransformationScale(scaling_factor)
Calculate the transformation scale for the imgae. The scale is a numpy array of size two indicating the width and height scale.
Arguments
- image: Numpy array
- size: Numpy array of length 2
GetSourceDestinationPoints
paz.processors.geometric.GetSourceDestinationPoints(scaling_factor)
Returns the source and destination points for affine transformation.
Arguments
- center: Numpy array of shape (2,). Center coordinates of image
- scale: Numpy array of shape (2,). Scale of width and height of image
- size: List of length 2. Size of image
GetImageCenter
paz.processors.geometric.GetImageCenter(offset=0.5)
Calculate the center of the image and add an offset to the center.
Arguments
- image: Numpy array
- offset: Float
WarpAffine
paz.processors.geometric.WarpAffine()
Applies an affine transformation to an image
Arguments
- image: Numpy array
- transform: Numpy array. Transformation matrix
- size: Numpy array. Transformation size