Geometric transforms (rotate, flip, crop) in Computer Vision - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of rotating an image by 90 degrees clockwise

Given a 3x3 grayscale image represented as a 2D list, what is the output after rotating it 90 degrees clockwise?

Computer Vision

import numpy as np
image = np.array([[1,2,3],[4,5,6],[7,8,9]])
rotated = np.rot90(image, k=-1)
print(rotated.tolist())

A [[7,4,1],[8,5,2],[9,6,3]]

B [[9,8,7],[6,5,4],[3,2,1]]

C [[3,6,9],[2,5,8],[1,4,7]]

D [[1,2,3],[4,5,6],[7,8,9]]

💡 Hint

Rotating 90 degrees clockwise means columns become rows from bottom to top.

The numpy rot90 function with k=-1 rotates the array 90 degrees clockwise. The first column becomes the first row reversed.

❓ Model Choice

intermediate

Choosing the correct transform for horizontal flip

Which numpy operation correctly flips a 2D image array horizontally?

A np.transpose(image)

B np.flipud(image)

C np.rot90(image, k=2)

D np.fliplr(image)

💡 Hint

Horizontal flip means flipping left to right.

np.fliplr flips the array left to right (horizontally). np.flipud flips up and down (vertical flip).

🔧 Debug

advanced

Identify the error in cropping code

What error will this code raise when cropping a 2D image array?

Computer Vision

import numpy as np
image = np.arange(16).reshape(4,4)
cropped = image[1:3, 3:1]
print(cropped)

A TypeError: slice indices must be integers or None or have an __index__ method

B IndexError: slice indices must be increasing

C Returns an empty array with shape (2,0)

D SyntaxError due to invalid slice

💡 Hint

Check how slicing with start index greater than stop index behaves in numpy.

Slicing with start > stop returns an empty slice, so cropped has shape (2,0) and prints []

❓ Hyperparameter

advanced

Effect of interpolation parameter in rotation

When rotating an image using OpenCV's cv2.warpAffine, which interpolation method preserves the most image quality for small rotations?

A cv2.INTER_CUBIC

B cv2.INTER_LINEAR

C cv2.INTER_AREA

D cv2.INTER_NEAREST

💡 Hint

Higher order interpolation methods produce smoother results.

cv2.INTER_CUBIC uses bicubic interpolation which preserves quality better than linear or nearest neighbor for rotations.

❓ Metrics

expert

Evaluating effect of geometric transforms on classification accuracy

You apply random rotations and flips as data augmentation during training a CNN classifier. After training, which metric change best indicates the augmentation improved model generalization?

A Training accuracy increases but validation accuracy decreases

B Training accuracy decreases but validation accuracy increases

C Both training and validation accuracy decrease

D Both training and validation accuracy increase

💡 Hint

Good augmentation reduces overfitting, so training accuracy may drop but validation improves.

Data augmentation makes training harder, lowering training accuracy, but helps generalization, raising validation accuracy.