Image augmentation transforms in Computer Vision

Introduction

Image augmentation transforms help create more varied pictures from a few images. This makes machine learning models better at understanding new pictures.

When you have a small number of images to train a model.

When you want your model to recognize objects from different angles or lighting.

When you want to reduce overfitting by showing the model many versions of the same image.

When you want to simulate real-world changes like rotation, flipping, or zooming.

When you want to improve model accuracy without collecting more data.

Syntax

Computer Vision

transform = torchvision.transforms.Compose([
    torchvision.transforms.RandomHorizontalFlip(p=0.5),
    torchvision.transforms.RandomRotation(degrees=30),
    torchvision.transforms.ColorJitter(brightness=0.2, contrast=0.2),
    torchvision.transforms.ToTensor()
])

Use Compose to chain multiple transforms together.

Each transform changes the image in a simple way, like flipping or rotating.

Examples

Always flips the image horizontally.

Computer Vision

transform = torchvision.transforms.RandomHorizontalFlip(p=1.0)

Rotates the image randomly within ±45 degrees.

Computer Vision

transform = torchvision.transforms.RandomRotation(degrees=45)

Randomly changes the brightness of the image.

Computer Vision

transform = torchvision.transforms.ColorJitter(brightness=0.5)

Randomly crops and resizes the image to 224x224 pixels, then converts it to a tensor.

Computer Vision

transform = torchvision.transforms.Compose([
    torchvision.transforms.RandomResizedCrop(224),
    torchvision.transforms.ToTensor()
])

Sample Model

This code loads an image from the internet, applies several augmentation transforms, and prints the shape and type of the resulting tensor. The image is flipped, rotated, brightness and contrast changed, then converted to a tensor.

Computer Vision

import torch
from torchvision import transforms
from PIL import Image
import requests
from io import BytesIO

# Load an example image from the web
url = 'https://upload.wikimedia.org/wikipedia/commons/thumb/4/47/PNG_transparency_demonstration_1.png/640px-PNG_transparency_demonstration_1.png'
response = requests.get(url)
img = Image.open(BytesIO(response.content))

# Define augmentation transforms
transform = transforms.Compose([
    transforms.RandomHorizontalFlip(p=1.0),
    transforms.RandomRotation(degrees=30),
    transforms.ColorJitter(brightness=0.3, contrast=0.3),
    transforms.ToTensor()
])

# Apply transforms
augmented_img = transform(img)

# Show shape and type
print(f"Augmented image tensor shape: {augmented_img.shape}")
print(f"Tensor type: {augmented_img.dtype}")

Important Notes

Transforms like RandomHorizontalFlip and RandomRotation add variety to training images.

Always convert images to tensors before feeding them to models.

Augmentation should be applied only to training data, not validation or test data.

Summary

Image augmentation creates new images by changing originals slightly.

Transforms include flipping, rotating, cropping, and color changes.

Using augmentation helps models learn better and avoid overfitting.