Image augmentation transforms in Computer Vision - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of horizontal flip augmentation

What will be the shape and pixel value of the center pixel after applying a horizontal flip to this 3x3 grayscale image?

Computer Vision

import numpy as np
image = np.array([[10, 20, 30],
                  [40, 50, 60],
                  [70, 80, 90]])
flipped = np.fliplr(image)
center_pixel = flipped[1,1]
print(flipped.shape, center_pixel)

A (3, 3) 40

B (3, 3) 80

C (3, 3) 60

D (3, 3) 50

💡 Hint

Horizontal flip reverses columns but keeps rows the same.

The horizontal flip reverses each row. The center pixel stays at position (1,1) and keeps its original value 50.

❓ Model Choice

intermediate

Best augmentation for rotation invariance

Which image augmentation transform helps a model learn rotation invariance best?

A Random rotation by angles between -30 and 30 degrees

B Random brightness adjustment

C Random horizontal flip

D Random cropping

💡 Hint

Rotation invariance means the model should recognize objects even if rotated.

Random rotation teaches the model to recognize objects at different angles, improving rotation invariance.

❓ Hyperparameter

advanced

Choosing brightness adjustment range

You want to augment images by adjusting brightness. Which brightness factor range is most reasonable to avoid unnatural images?

A [0.0, 3.0]

B [0.1, 2.0]

C [0.5, 1.5]

D [1.0, 5.0]

💡 Hint

Brightness factor 1 means no change; too low or too high values can distort images.

A range from 0.5 to 1.5 adjusts brightness moderately, keeping images natural while augmenting.

❓ Metrics

advanced

Effect of augmentation on validation accuracy

After adding random cropping augmentation during training, which effect on validation accuracy is most likely?

A Validation accuracy increases due to better generalization

B Validation accuracy fluctuates randomly without pattern

C Validation accuracy stays the same

D Validation accuracy decreases due to overfitting

💡 Hint

Augmentation usually helps models generalize better to unseen data.

Random cropping creates varied training samples, helping the model generalize and improving validation accuracy.

🔧 Debug

expert

Bug in augmentation pipeline code

What error will this code raise when applying a vertical flip using torchvision transforms on a PIL image?

Computer Vision

from torchvision import transforms
from PIL import Image

image = Image.new('RGB', (100, 100), color='red')
transform = transforms.RandomVerticalFlip(p=1.0)
augmented = transform(image)
print(type(augmented))

A AttributeError: 'RandomVerticalFlip' object has no attribute 'forward'

B No error, prints <class 'PIL.Image.Image'>

C TypeError: 'RandomVerticalFlip' object is not callable

D RuntimeError: transform requires tensor input

💡 Hint

Check if RandomVerticalFlip works directly on PIL images.

RandomVerticalFlip from torchvision.transforms works on PIL images and returns a PIL image, so no error occurs.