tfm.vision.augment.RandomErasing

Applies RandomErasing to a single image.

Inherits From: ImageAugment

tfm.vision.augment.RandomErasing(
    probability: float = 0.25,
    min_area: float = 0.02,
    max_area: float = (1 / 3),
    min_aspect: float = 0.3,
    max_aspect: Optional[float] = None,
    min_count=1,
    max_count=1,
    trials=10
)

Reference: https://arxiv.org/abs/1708.04896

Implementation is inspired by https://github.com/rwightman/pytorch-image-models

probability Probability of augmenting the image. Defaults to 0.25.
min_area Minimum area of the random erasing rectangle. Defaults to 0.02.
max_area Maximum area of the random erasing rectangle. Defaults to 1/3.
min_aspect Minimum aspect rate of the random erasing rectangle. Defaults to 0.3.
max_aspect Maximum aspect rate of the random erasing rectangle. Defaults to None.
min_count Minimum number of erased rectangles. Defaults to 1.
max_count Maximum number of erased rectangles. Defaults to 1.
trials Maximum number of trials to randomly sample a rectangle that fulfills constraint. Defaults to 10.

Methods

distort

View source

distort(
    image: tf.Tensor
) -> tf.Tensor

Applies RandomErasing to single image.

Args
image tf.Tensor

Of shape [height, width, 3] representing an image.

Returns
tf.Tensor The augmented version of image.

distort_with_boxes

View source

distort_with_boxes(
    image: tf.Tensor, bboxes: tf.Tensor
) -> Tuple[tf.Tensor, tf.Tensor]

Distorts the image and bounding boxes.

Expect the image tensor values are in the range [0, 255].

Args
image Tensor of shape [height, width, 3] or [num_frames, height, width, 3] representing an image or image sequence.
bboxes Tensor of shape [num_boxes, 4] or [num_frames, num_boxes, 4] representing bounding boxes for an image or image sequence.

Returns
The augmented version of image and bboxes.