CS5720 - Data Augmentation for Images

What is Data Augmentation?

Data augmentation artificially expands training datasets by applying transformations to existing images, creating new training examples without collecting additional data.

Increases dataset diversity by creating variations of existing images
Improves model generalization by exposing it to more data variations
Reduces overfitting by providing more training examples
Cost-effective alternative to collecting more labeled data
Preserves labels while creating semantically meaningful variations

Key Benefits

🛡️ Improved Robustness

Models become more resilient to variations in lighting, orientation, and scale

🎯 Better Generalization

Reduced overfitting leads to better performance on unseen test data

🔄 Translation Invariance

Models learn to recognize objects regardless of position or orientation

📈 Data Efficiency

Achieve better results with smaller original datasets

⚖️ Class Balance

Address class imbalance by augmenting underrepresented classes

Common Augmentation Techniques

🔄

Rotation

Rotate images by random angles to simulate different viewpoints

🪞

Flipping

Horizontal/vertical flips for left-right symmetry

🔍

Scaling

Zoom in/out to simulate different distances

📐

Translation

Shift images to simulate camera movement

💡

Brightness

Adjust brightness for different lighting conditions

⚫

Contrast

Modify contrast to simulate exposure variations

📶

Noise

Add random noise to simulate sensor noise

🌫️

Blur

Apply blur effects to simulate motion or focus issues

Implementation Example

import torch
from torchvision import transforms

# Define augmentation pipeline
train_transform = transforms.Compose([
    transforms.RandomRotation(degrees=30),
    transforms.RandomHorizontalFlip(p=0.5),
    transforms.RandomResizedCrop(size=224, scale=(0.8, 1.0)),
    transforms.ColorJitter(brightness=0.2, contrast=0.2),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], 
                        std=[0.229, 0.224, 0.225])
])

# Apply to dataset
augmented_dataset = ImageFolder(root='data/train', 
                               transform=train_transform)
            

🎛️ Use Moderation

Avoid over-augmentation that distorts images beyond recognition

🎯 Domain-Specific

Choose augmentations appropriate for your specific problem domain

📊 Validation Strategy

Apply minimal or no augmentation to validation sets

📈 Progressive Approach

Start with simple augmentations and gradually add complexity

Data Augmentation for Images

What is Data Augmentation?

Key Benefits

Common Augmentation Techniques

Implementation Example

Modal Title