Data Augmentation¶

Data augmentation is crucial for PhenoCLR training, creating diverse views of the same image to learn robust representations. Phenoscape uses the Kornia library for augmentations across RGB, multispectral, and hyperspectral data.

Overview¶

PhenoCLR requires two augmented views of each image to create positive pairs for contrastive learning. The quality and diversity of augmentations directly impact the learned representations' robustness and biological relevance.

Augmentation Categories¶

Geometric Transformations¶

Augmentation	Parameters	Biological Relevance
`RandomResizedCrop`	`scale=(0.08, 1.0)`	Simulates different viewing distances
`RandomHorizontalFlip`	`p=0.5`	Natural left-right symmetry
`RandomVerticalFlip`	`p=0.2`	Less common but valid orientation
`RandomRotation`	`degrees=15`	Natural pose variations
`RandomPerspective`	`distortion_scale=0.2`	Camera angle variations

Color Transformations¶

Augmentation	Parameters	Purpose
`ColorJitter`	`brightness=0.4, contrast=0.4`	Lighting condition variations
`RandomGrayscale`	`p=0.2`	Forces learning of shape features
`RandomPosterize`	`bits=3`	Reduces color depth
`RandomSharpness`	`sharpness=0.5`	Image quality variations

Noise and Distortions¶

Augmentation	Parameters	Effect
`RandomGaussianBlur`	`kernel_size=(3,3), sigma=(0.1,2.0)`	Motion blur, focus variations
`RandomErase`	`scale=(0.02,0.33)`	Occlusion simulation
`RandomThinPlateSpline`	`scale=0.2`	Elastic deformations

Configuration¶

YAML Configuration¶

# Augmentation configuration
augmentations:
  # Geometric
  random_crop:
    size: 224
    scale: [0.08, 1.0]
    ratio: [0.75, 1.33]

  random_flip:
    horizontal: 0.5
    vertical: 0.2

  random_rotation:
    degrees: 15
    probability: 0.7

  # Color
  color_jitter:
    brightness: 0.4
    contrast: 0.4
    saturation: 0.4
    hue: 0.1
    probability: 0.8

  random_grayscale:
    probability: 0.2

  # Blur and noise
  gaussian_blur:
    kernel_size: [3, 3]
    sigma: [0.1, 2.0]
    probability: 0.5

  # Advanced
  random_erase:
    scale: [0.02, 0.33]
    ratio: [0.3, 3.3]
    probability: 0.5

Programmatic Configuration¶

def create_augmentation_pipeline(config):
    """Create augmentation pipeline from configuration."""
    transforms = []

    # Geometric transforms
    if 'random_crop' in config:
        crop_config = config['random_crop']
        transforms.append(
            K.RandomResizedCrop(
                size=crop_config['size'],
                scale=crop_config['scale'],
                ratio=crop_config['ratio']
            )
        )

    if 'random_flip' in config:
        flip_config = config['random_flip']
        transforms.append(K.RandomHorizontalFlip(p=flip_config['horizontal']))
        transforms.append(K.RandomVerticalFlip(p=flip_config['vertical']))

    # Color transforms
    if 'color_jitter' in config:
        cj_config = config['color_jitter']
        transforms.append(
            K.ColorJitter(
                brightness=cj_config['brightness'],
                contrast=cj_config['contrast'],
                saturation=cj_config['saturation'],
                hue=cj_config['hue'],
                p=cj_config['probability']
            )
        )

    return K.AugmentationSequential(*transforms, same_on_batch=False)