basecls.data.rand_erase#

Random Erasing (Cutout)

Random Erasing: “Random Erasing Data Augmentation”

引用

rwightman/pytorch-image-models

class basecls.data.rand_erase.RandomErasing(prob=0.5, scale_range=(0.02, 1 / 3), ratio=0.3, mode='const', count=1, num_splits=0, pad_mean=0.0, pad_std=1.0, *, order=None)[源代码]#

基类：VisionTransform

Randomly selects a rectangle region in an image and erases its pixels.

This variant of RandomErasing is intended to be applied to either a batch or single image tensor after it has been normalized by dataset mean and std.

参数

  • prob (float) – probability that Random Erasing operation will be performed. Default: 0.5

  • scale_range (Tuple[float, float]) – percentage of erased area wrt input image area. Default: (0.02, 1.0 / 3)

  • ratio (Union[float, Tuple[float, float]]) – aspect ratio of erased area. if a scalar, range will be (ratio, 1.0 / ratio). Default: 0.3

  • mode (str) – pixel color mode, one of “const”, “rand”, or “pixel”. Default: "const" "const" - erase block is constant color of 0 for all channels "rand" - erase block is same per-channel random (normal) color "pixel" - erase block is per-pixel random (normal) color

  • count (Union[int, Tuple[int, int]]) – maximum number or range of erasing blocks per image, area per box is scaled by count. if a scalar, per-image count is randomly chosen between 1 and this value. if a range, per-image count is randomly chosen between this range. Default: 1

  • num_splits (int) – augmentation splits. if > 1, the first split will not be erased. Default: 0

  • pad_mean (Union[float, Tuple[float, float, float]]) – the mean of padding pixels. Default: 0.0

  • pad_std (Union[float, Tuple[float, float, float]]) – the std of padding pixels. Default: 1.0

apply_batch(inputs)[源代码]#

Apply transform on batch input data.