Computer vision basics in Intro to Computing - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the primary purpose of image preprocessing in computer vision?

Image preprocessing is a crucial step before feeding images into a model. What is its main goal?

A To enhance image quality and normalize data for better model performance

B To increase the image file size for better resolution

C To convert images into audio signals for analysis

D To randomly shuffle pixel values to create data augmentation

💡 Hint

Think about how preparing data helps a model learn better.

Image preprocessing improves image quality and standardizes input data, making it easier for models to learn patterns effectively.

🔍 Analysis

intermediate

What is the output shape after applying a 3x3 convolution with stride 1 and padding 1 on a 28x28 grayscale image?

Given a single grayscale image of size 28x28 pixels, a convolutional layer uses a 3x3 filter, stride 1, and padding 1. What will be the output shape?

Intro to Computing

import torch
import torch.nn as nn

input_tensor = torch.randn(1, 1, 28, 28)  # batch_size=1, channels=1, height=28, width=28
conv = nn.Conv2d(in_channels=1, out_channels=10, kernel_size=3, stride=1, padding=1)
output = conv(input_tensor)
output.shape

A torch.Size([1, 10, 26, 26])

B torch.Size([1, 10, 28, 28])

C torch.Size([1, 1, 28, 28])

D torch.Size([1, 10, 30, 30])

💡 Hint

Padding of 1 keeps the spatial size same when stride is 1 and kernel size is 3.

Padding of 1 pixel on each side with a 3x3 kernel and stride 1 keeps the height and width unchanged. The output channels equal the number of filters (10).

❓ Model Choice

advanced

Which model architecture is best suited for detecting objects at multiple scales in an image?

You want to build a computer vision model that detects objects of various sizes in images. Which architecture is most appropriate?

A A simple fully connected neural network

B A recurrent neural network (RNN)

C A convolutional neural network with feature pyramid networks (FPN)

D A linear regression model

💡 Hint

Think about architectures designed to handle different object sizes.

Feature Pyramid Networks (FPN) enhance CNNs by combining features at multiple scales, improving detection of objects of different sizes.

❓ Metrics

advanced

Which metric is most appropriate to evaluate a binary image segmentation model?

You trained a model to segment objects in images (pixel-wise classification). Which metric best measures overlap between predicted and true segments?

A Mean Squared Error (MSE)

B Precision

C Accuracy

D Intersection over Union (IoU)

💡 Hint

This metric compares the area of overlap to the area of union between prediction and ground truth.

IoU measures how well the predicted segmentation overlaps with the true segmentation, making it ideal for segmentation tasks.

🔍 Analysis

expert

Why does this CNN training code raise a runtime error?

Examine the code below. It raises a runtime error during training. What is the cause?

Intro to Computing

import torch
import torch.nn as nn

class SimpleCNN(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv = nn.Conv2d(3, 16, 3)
        self.fc = nn.Linear(16*30*30, 10)
    def forward(self, x):
        x = self.conv(x)
        x = torch.relu(x)
        x = x.view(-1, 16*30*30)
        x = self.fc(x)
        return x

model = SimpleCNN()
input_tensor = torch.randn(4, 3, 32, 32)
output = model(input_tensor)

A The input to the linear layer has incorrect size due to convolution output shape mismatch

B The convolution layer has wrong number of input channels

C The ReLU function is applied incorrectly

D The batch size is too large for the model

💡 Hint

Check the size of the tensor after convolution before flattening.

Without padding, the 3x3 convolution reduces spatial dimensions from 32x32 to 30x30 ((32-3+0)/1 +1 = 30). The view flattens to 16*30*30 = 14400 features, but the Linear layer expects 16*30*30 = 14400, causing a shape mismatch runtime error.