Python CV ecosystem (OpenCV, PIL, torchvision) in Computer Vision - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

What is the output of this OpenCV image shape code?

Given an image loaded with OpenCV, what will be the output of the shape attribute?

Computer Vision

import cv2
img = cv2.imread('sample.jpg')
print(img.shape)

A (height, width, channels) tuple representing image dimensions

B (width, height, channels) tuple representing image dimensions

C A single integer representing total pixels

D Raises an error because shape attribute does not exist

💡 Hint

OpenCV loads images as NumPy arrays with shape (rows, columns, channels).

OpenCV loads images as NumPy arrays with shape (height, width, channels). The shape attribute returns a tuple in this order.

❓ Model Choice

intermediate

Which torchvision model is best for image classification on ImageNet?

You want to use a pretrained model from torchvision for classifying images into 1000 classes. Which model is designed specifically for this?

A torchvision.models.segmentation.fcn_resnet50(pretrained=True)

B torchvision.models.resnet50(pretrained=True)

C torchvision.models.detection.fasterrcnn_resnet50_fpn(pretrained=True)

D torchvision.models.video.r3d_18(pretrained=True)

💡 Hint

Image classification models output class probabilities for fixed classes.

ResNet50 is a standard image classification model pretrained on ImageNet with 1000 classes. The others are for segmentation, detection, or video tasks.

❓ Metrics

advanced

How to correctly compute accuracy for a torchvision classification model?

You have model outputs as logits and true labels as integers. Which code snippet correctly computes accuracy?

A

preds = outputs.argmax(dim=1)
accuracy = (preds == labels).sum().item()

B

preds = outputs.max(dim=0)
accuracy = (preds == labels).float().mean().item()

C

preds = outputs.argmax(dim=1)
accuracy = (preds == labels).float().mean().item()

D

preds = outputs.argmax(dim=1)
accuracy = (preds == labels).float().sum().item() / len(labels)

💡 Hint

Accuracy is the fraction of correct predictions over total samples.

Option C correctly finds predicted classes by argmax over dimension 1, compares with labels, converts to float, averages, and extracts scalar.

🔧 Debug

advanced

Why does this PIL image conversion code raise an error?

Code snippet: from PIL import Image img = Image.open('photo.png') img = img.convert('HSV')

Computer Vision

from PIL import Image
img = Image.open('photo.png')
img = img.convert('HSV')

A No error, image converts successfully

B FileNotFoundError because 'photo.png' does not exist

C ValueError because 'HSV' is not a supported mode in PIL

D TypeError because convert expects an integer

💡 Hint

PIL supports specific modes like 'RGB', 'L', 'RGBA'.

The convert method accepts 'HSV' as a valid mode in PIL, so no error is raised.

🧠 Conceptual

expert

What is the main difference between OpenCV and PIL in image processing?

Choose the statement that best describes a key difference between OpenCV and PIL libraries.

A PIL is designed for real-time computer vision, OpenCV is not

B PIL supports video processing, OpenCV does not

C OpenCV cannot read PNG images, PIL can

D OpenCV uses BGR color order by default, PIL uses RGB

💡 Hint

Think about default color channel order in each library.

OpenCV loads images in BGR order by default, while PIL uses RGB order. This often causes color confusion if not handled.