Unified Image Augmentation Pipeline

# Illustrative sketch (not from the paper)
import albumentations as A, numpy as np

# Compose bundles transforms, probabilities, annotation rules, and a seed
transform = A.Compose(
    [A.RandomCrop(width=256, height=256, p=1.0),
     A.HorizontalFlip(p=0.5),
     A.Rotate(limit=15, p=0.7)],
    bbox_params=A.BboxParams(format='coco', label_fields=['labels']),
    keypoint_params=A.KeypointParams(format='xy'),
    p=1.0,
    seed=123
)

sample = {
    "image": np.zeros((512, 512, 3), dtype=np.uint8),
    "mask": np.zeros((512, 512), dtype=np.uint8),
    "bboxes": [[100, 120, 200, 220]],
    "labels": [1],
    "keypoints": [(150, 160)]
}

# One call generates random values applied to all targets
out = transform(**sample)

// Illustrative sketch (not from the paper)
const A = require('albumentationsx');
const { randomCrop, horizontalFlip, rotate } = A;

// Compose with transforms, annotation handling, and a fixed seed
const transform = new A.Compose({
  transforms: [
    randomCrop({ width: 256, height: 256, p: 1.0 }),
    horizontalFlip({ p: 0.5 }),
    rotate({ limit: 15, p: 0.7 })
  ],
  bboxParams: { format: 'coco', labelFields: ['labels'] },
  keypointParams: { format: 'xy' },
  p: 1.0,
  seed: 123
});

const sample = {
  image: Buffer.alloc(512 * 512 * 3), // placeholder image data
  mask: Buffer.alloc(512 * 512),
  bboxes: [[100, 120, 200, 220]],
  labels: [1],
  keypoints: [[150, 160]]
};

// Single invocation applies identical random decisions to all targets
const out = transform.apply(sample);