albumentations.augmentations.dropout.coarse_dropout

CoarseDropout randomly drops out rectangular regions from the image and optionally, the corresponding regions in an associated mask, to simulate occlusion and varied object sizes found in real-world settings. This transformation is an evolution of CutOut and RandomErasing, offering more flexibility in the size, number of dropout regions, and fill values.

Parameters

Example

>>> 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 random uniform fill
>>> aug_random = A.CoarseDropout(
...     num_holes_range=(3, 6),
...     hole_height_range=(10, 20),
...     hole_width_range=(10, 20),
...     fill="random_uniform",
...     p=1.0
... )
>>> # Example with inpainting
>>> aug_inpaint = A.CoarseDropout(
...     num_holes_range=(3, 6),
...     hole_height_range=(10, 20),
...     hole_width_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

- The actual number and size of dropout regions are randomly chosen within the specified ranges for each application. - When using float values for hole_height_range and hole_width_range, ensure they are between 0 and 1. - This implementation includes deprecation warnings for older parameter names (min_holes, max_holes, etc.). - Inpainting methods ('inpaint_telea', 'inpaint_ns') work only with grayscale or RGB images. - For 'random_uniform' fill, each hole gets a single random color, unlike 'random' where each pixel gets its own random value.

References

• CutOut: https://arxiv.org/abs/1708.04552
• Random Erasing: https://arxiv.org/abs/1708.04896
• OpenCV Inpainting methods: https://docs.opencv.org/master/df/d3d/tutorial_py_inpainting.html

Applies coarse dropout to regions containing specific objects in the image. This augmentation creates holes (dropout regions) for each target object in the image. Objects can be specified either by their class indices in a segmentation mask or by their labels in bounding box annotations. The hole generation differs between mask and box modes: Mask mode: 1. For each connected component in the mask matching target indices: - Samples N points randomly from within the object region (with replacement) - Creates holes centered at these points - Hole sizes are proportional to sqrt(component area), not total object area - Each component's holes are sized based on its own area Box mode: 1. For each bounding box matching target labels: - Creates N holes with random positions inside the box - Hole sizes are proportional to the box dimensions In both modes: - N is sampled once from num_holes_range and used for all objects - For example, if num_holes_range=(2,4) and 3 is sampled: * With 3 target objects, you'll get exactly 3 holes per object (9 total) * Holes may overlap within or between objects * All holes are clipped to image boundaries

Parameters

Notes

At least one of mask_indices or bbox_labels must be provided. If both are provided, mask_indices takes precedence.

Randomly erases rectangular regions in an image, following the Random Erasing Data Augmentation technique. This augmentation helps improve model robustness by randomly masking out rectangular regions in the image, simulating occlusions and encouraging the model to learn from partial information. It's particularly effective for image classification and person re-identification tasks.

Parameters

Example

>>> import numpy as np
>>> import albumentations as A
>>> image = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8)
>>> # Basic usage with default parameters
>>> transform = A.Erasing()
>>> transformed = transform(image=image)
>>> # Custom configuration
>>> transform = A.Erasing(
...     scale=(0.1, 0.4),
...     ratio=(0.5, 2.0),
...     fill_value="random_uniform",
...     p=1.0
... )
>>> transformed = transform(image=image)

Notes

- The transform attempts to find valid erasing parameters up to 10 times. If unsuccessful, no erasing is performed. - The actual erased area and aspect ratio are randomly sampled within the specified ranges for each application. - When using inpainting methods, only grayscale or RGB images are supported.

References

• Paper: https://arxiv.org/abs/1708.04896
• Implementation inspired by torchvision: https://pytorch.org/vision/stable/transforms.html#torchvision.transforms.RandomErasing