Albumentations

图像增强库首选！ 官网: https://albumentations.ai/

pip uninstall albumentations  # If you have it installed
pip install albumentationsx   # Install the new version

如果是RGBA 图像增强需要注意选择通道属性

https://albumentations.ai/docs/2-core-concepts/pipelines/#selectivechanneltransform-apply-transforms-to-specific-channels

import albumentations as A
import numpy as np

# Assume a 4-channel image (RGBA)
image_rgba = np.random.randint(0, 256, (100, 100, 4), dtype=np.uint8)

# Apply GaussianBlur only to the first three channels (RGB)
pipeline = A.Compose([
    A.SelectiveChannelTransform(
        A.GaussianBlur(p=1.0), # Transform to apply
        channels=[0, 1, 2], # Apply only to RGB channels
        p=1.0 # Probability of applying this selective transform
    ),
    A.HorizontalFlip(p=0.5) # Applied to all channels
])

transformed_data = pipeline(image=image_rgba)
transformed_image = transformed_data['image']

# When pipeline(image=...) is called:
# - GaussianBlur is applied *only* to the R, G, B channels.
# - Independently, 50% chance the *entire* RGBA image is flipped.