Albumentations

Albumentations is a high-performance Python library for image augmentation designed for deep learning and computer vision tasks.

Key features include:

• Complete CV Support: Works with classification, segmentation (semantic & instance), object detection, and pose estimation.
• Unified API: Consistent interface for RGB/grayscale/multispectral images, masks, bounding boxes, and keypoints.
• Rich Library: Over 70 high-quality augmentations.
• Fast Execution: Optimized for production use and benchmarked as one of the fastest augmentation libraries.
• Framework Integration: Compatible with PyTorch, TensorFlow, and other major deep learning frameworks.

Pixel-level transforms modify only the input image. They do not affect additional targets like masks, bounding boxes, or keypoints.

For volumetric data (volumes and 3D masks), these transforms are applied independently to each slice along the Z-axis (depth dimension) to maintain consistency across the volume.

Spatial-level transforms modify both the input image and any associated additional targets (e.g., masks, bounding boxes, and keypoints) simultaneously.

For volumetric data, these transforms are applied independently to each slice along the Z-axis (depth dimension) to maintain consistency across the volume.

Albumentations requires Python 3.9 or higher. You can install the latest version from PyPI using pip.

pip install -U albumentations

The original albumentations repository is no longer actively maintained. For ongoing support, bug fixes, and performance improvements, you should migrate to AlbumentationsX.

AlbumentationsX is a drop-in replacement with the same API. You can migrate by uninstalling the original package and installing albumentationsx. Your existing code using import albumentations as A will continue to work without changes.

Licensing Note: AlbumentationsX uses dual licensing (AGPL-3.0 / Commercial). The AGPL-3.0 license is not compatible with permissive licenses like MIT, Apache 2.0, or BSD. If your project uses these licenses, a commercial license is required.

# Uninstall original
pip uninstall albumentations

# Install AlbumentationsX
pip install albumentationsx

Albumentations provides a unified API to create augmentation pipelines using A.Compose. You can apply various transforms (like cropping, flipping, or brightness adjustments) to an image. The pipeline returns a dictionary containing the transformed image and any other augmented targets.

import albumentations as A
import cv2

# Declare an augmentation pipeline
transform = A.Compose([
    A.RandomCrop(width=256, height=256),
    A.HorizontalFlip(p=0.5),
    A.RandomBrightnessContrast(p=0.2),
])

# Read an image with OpenCV and convert it to the RGB colorspace
image = cv2.imread("image.jpg")
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

# Augment an image
transformed = transform(image=image)
transformed_image = transformed["image"]

By default, Albumentations performs a version check on initialization to notify users of updates. You can disable this behavior by setting the NO_ALBUMENTATIONS_UPDATE environment variable to true or 1.

# Disable update checks in your shell environment
export NO_ALBUMENTATIONS_UPDATE=true

3D transforms operate on volumetric data and can modify both the input volume and associated 3D mask.

Target Definitions:

• Volume: 3D array of shape (D, H, W) or (D, H, W, C).
• Mask3D: Binary or multi-class 3D mask of shape (D, H, W).

| Transform | Volume | Mask3D | Keypoints |
| :--- | :---: | :---: | :---: |
| CenterCrop3D | ✓ | ✓ | ✓ |
| CoarseDropout3D | ✓ | ✓ | ✓ |
| CubicSymmetry | ✓ | ✓ | ✓ |
| Pad3D | ✓ | ✓ | ✓ |
| PadIfNeeded3D | ✓ | ✓ | ✓ |
| RandomCrop3D | ✓ | ✓ | ✓ |

If you use Albumentations in your research, please cite the following BibTeX entry:

@Article{info11020125,
    AUTHOR = {Buslaev, Alexander and Iglovikov, Vladimir I. and Khvedchenya, Eugene and Parinov, Alex and Druzhinin, Mikhail and Kalinin, Alexandr A.},
    TITLE = {Albumentations: Fast and Flexible Image Augmentations},
    JOURNAL = {Information},
    VOLUME = {11},
    YEAR = {2020},
    NUMBER = {2},
    ARTICLE-NUMBER = {125},
    URL = {https://www.mdpi.com/2078-2489/11/2/125},
    ISSN = {2078-2489},
    DOI = {10.3390/info11020125}
}