Albucore is a library of optimized atomic functions designed for efficient image processing. These functions serve as the foundation for AlbumentationsX, a popular image augmentation library.

Image processing operations can be implemented in various ways, each with its own performance characteristics depending on the image type, size, and number of channels. Albucore aims to provide the fastest implementation for each operation by leveraging different backends such as NumPy, OpenCV, and custom optimized code.

Supported dtypes: uint8 and float32 only.

Key features:

• Optimized atomic image processing functions
• Automatic selection of the fastest implementation based on input image characteristics
• Seamless integration with AlbumentationsX
• Benchmark scripts available (see ./benchmark.sh <data_dir>)

Requires Python 3.10+. Basic installation (you manage OpenCV separately):

pip install albucore

With OpenCV headless (recommended for servers/CI):

pip install albucore[headless]

With OpenCV GUI support (for local development with cv2.imshow):

pip install albucore[gui]

With OpenCV contrib modules:

pip install albucore[contrib]              # GUI version
pip install albucore[contrib-headless]     # Headless version

Note: If you already have opencv-python or opencv-contrib-python installed, just use pip install albucore to avoid package conflicts. Albucore will detect and use your existing OpenCV installation.

import numpy as np
import albucore

# Create a sample RGB image
image = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8)

# Apply a function
result = albucore.multiply(image, 1.5)

# For grayscale images, ensure the channel dimension is present
gray_image = np.random.randint(0, 256, (100, 100, 1), dtype=np.uint8)
gray_result = albucore.multiply(gray_image, 1.5)

Albucore automatically selects the most efficient implementation based on the input image type and characteristics.

Albucore expects images to follow specific shape conventions, with the channel dimension always present:

• Single image: (H, W, C) - Height, Width, Channels
• Grayscale image: (H, W, 1) - Height, Width, 1 channel
• Batch of images: (N, H, W, C) - Number of images, Height, Width, Channels
• 3D volume: (D, H, W, C) - Depth, Height, Width, Channels
• Batch of volumes: (N, D, H, W, C) - Number of volumes, Depth, Height, Width, Channels

1. Channel dimension is always required, even for grayscale images (use shape (H, W, 1))
2. Single-channel images should have shape (H, W, 1) not (H, W)
3. Batch vs volume: (N, H, W, C) is N separate images; a single 3D volume is (D, H, W, C) with depth D. Do not confuse N (batch) with D (slices).

import numpy as np
import albucore

# Grayscale image - MUST have explicit channel dimension
gray_image = np.random.randint(0, 256, (100, 100, 1), dtype=np.uint8)

# RGB image
rgb_image = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8)

# Batch of 10 grayscale images
batch_gray = np.random.randint(0, 256, (10, 100, 100, 1), dtype=np.uint8)

# 3D volume with 20 slices
volume = np.random.randint(0, 256, (20, 100, 100, 1), dtype=np.uint8)

# Batch of 5 RGB volumes, each with 20 slices
batch_volumes = np.random.randint(0, 256, (5, 20, 100, 100, 3), dtype=np.uint8)

All symbols below are exported via from albucore import * (or from albucore.functions import …). Every function accepts uint8 and float32 images; inputs must have an explicit channel dimension ((H, W, C) — never bare (H, W)).

value / factor / exponent can be a scalar, a length-C 1-D array (per-channel), or a full image-shaped array.

normalize is for fixed per-channel constants (ImageNet-style). normalize_per_image estimates stats from the image at call time.

axis accepts None/"global" (scalar), "per_channel" (shape (C,)), or any NumPy-style int/tuple[int, ...].

See docs/decorators.md for @preserve_channel_dim, @contiguous, @clipped, and @batch_transform (used internally, not re-exported).

Many functions in Albucore support batch processing out of the box. The library automatically handles different input shapes:

• Single images: (H, W, C)
• Batches: (N, H, W, C)
• Volumes: (D, H, W, C)
• Batch of volumes: (N, D, H, W, C)

Functions will preserve the input shape structure, applying operations efficiently across all images/slices in the batch.

See docs/decorators.md for internal decorator documentation (@preserve_channel_dim, @contiguous, @clipped, @batch_transform).

Albucore uses a combination of techniques to achieve high performance:

1. Multiple Implementations: Each function may have several implementations using different backends (NumPy, OpenCV, custom code).
2. Automatic Selection: The library automatically chooses the fastest implementation based on the input image type, size, and number of channels.
3. Optimized Algorithms: Custom implementations are optimized for specific use cases, often outperforming general-purpose libraries.
4. NumKong: SIMD blend for add_weighted; cdist for small pairwise_distances_squared; wide-accumulator moments for uint8 global mean/std in stats; scale for float32 multiply_by_constant (see docs/numkong-performance.md).

Micro-benchmarks vs NumPy/OpenCV: see benchmarks/README.md; quick run: uv run python benchmarks/benchmark_numkong.py (OpenCV rows are skipped if cv2 is not installed).

See docs/performance-optimization.md for detailed performance guidelines and best practices.

• AGENTS.md - AI development guidelines for working with this codebase
• docs/image-conventions.md - Image shape conventions and requirements
• docs/decorators.md - Decorator usage and patterns
• docs/performance-optimization.md - Performance optimization guidelines
• docs/numkong-performance.md - NumKong vs OpenCV/NumPy/LUT baselines (benchmark tables; sum/mean/std)
• docs/public-api.md - Star-exported routers vs albucore.functions shims
• benchmarks/README.md - Python micro-benchmarks (uv run python benchmarks/…)
• docs/research/ - Research notes (extra benchmark writeups; see benchmarks/README.md for scripts)

MIT

Albucore is part of the AlbumentationsX project. We'd like to thank all contributors to AlbumentationsX and the broader computer vision community for their inspiration and support.