...

GridDropout

Targets:
image
mask
bboxes
keypoints
volume
mask3d
Image Types:uint8, float32

Apply GridDropout augmentation to images, masks, bounding boxes, and keypoints.

GridDropout drops out rectangular regions of an image and the corresponding mask in a grid fashion. This technique can help improve model robustness by forcing the network to rely on a broader context rather than specific local features.

Arguments
ratio
float
0.5

The ratio of the mask holes to the unit size (same for horizontal and vertical directions). Must be between 0 and 1. Default: 0.5.

unit_size_range
tuple[int, int] | None

Range from which to sample grid size. Default: None. Must be between 2 and the image's shorter edge. If None, grid size is calculated based on image size.

holes_number_xy
tuple[int, int] | None

The number of grid units in x and y directions. First value should be between 1 and image width//2, Second value should be between 1 and image height//2. Default: None. If provided, overrides unit_size_range.

random_offset
bool
true

Whether to offset the grid randomly between 0 and (grid unit size - hole size). If True, entered shift_xy is ignored and set randomly. Default: True.

fill
tuple[float, ...] | float | random | random_uniform | inpaint_telea | inpaint_ns
0

Value for the dropped pixels. Can be:

  • int or float: all channels are filled with this value
  • tuple: tuple of values for each channel
  • 'random': each pixel is filled with random values
  • 'random_uniform': each hole is filled with a single random color
  • 'inpaint_telea': uses OpenCV Telea inpainting method
  • 'inpaint_ns': uses OpenCV Navier-Stokes inpainting method Default: 0
fill_mask
tuple[float, ...] | float | None

Value for the dropped pixels in mask. If None, the mask is not modified. Default: None.

shift_xy
tuple[int, int]
[0,0]

Offsets of the grid start in x and y directions from (0,0) coordinate. Only used when random_offset is False. Default: (0, 0).

p
float
0.5

Probability of applying the transform. Default: 0.5.

Examples
>>> import numpy as np
>>> import albumentations as A
>>> image = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8)
>>> mask = np.random.randint(0, 2, (100, 100), dtype=np.uint8)
>>> # Example with standard fill value
>>> aug_basic = A.GridDropout(
...     ratio=0.3,
...     unit_size_range=(10, 20),
...     random_offset=True,
...     p=1.0
... )
>>> # Example with random uniform fill
>>> aug_random = A.GridDropout(
...     ratio=0.3,
...     unit_size_range=(10, 20),
...     fill="random_uniform",
...     p=1.0
... )
>>> # Example with inpainting
>>> aug_inpaint = A.GridDropout(
...     ratio=0.3,
...     unit_size_range=(10, 20),
...     fill="inpaint_ns",
...     p=1.0
... )
>>> transformed = aug_random(image=image, mask=mask)
>>> transformed_image, transformed_mask = transformed["image"], transformed["mask"]
Notes
  • If both unit_size_range and holes_number_xy are None, the grid size is calculated based on the image size.
  • The actual number of dropped regions may differ slightly from holes_number_xy due to rounding.
  • Inpainting methods ('inpaint_telea', 'inpaint_ns') work only with grayscale or RGB images.
  • For 'random_uniform' fill, each grid cell gets a single random color, unlike 'random' where each pixel gets its own random value.