Image inpainting concept in Computer Vision - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the main goal of image inpainting?

Imagine you have a photo with some parts missing or damaged. What does image inpainting try to do with that photo?

A Detect objects in the photo and label them.

B Increase the resolution of the entire photo to make it clearer.

C Change the colors of the photo to make it look artistic.

D Fill in the missing or damaged parts with plausible content that matches the surrounding area.

💡 Hint

Think about fixing or restoring parts of an image.

Image inpainting is about filling missing or damaged areas in an image with content that looks natural and consistent with the rest of the image.

❓ Model Choice

intermediate

Which type of model is commonly used for image inpainting?

You want to build a model that fills missing parts of images realistically. Which model type is best suited for this?

A Convolutional Neural Network (CNN) with encoder-decoder architecture

B Recurrent Neural Network (RNN) for sequence prediction

C Support Vector Machine (SVM) for classification

D K-Nearest Neighbors (KNN) for clustering

💡 Hint

Think about models good at understanding images and generating pixels.

CNNs with encoder-decoder structures are widely used for image inpainting because they can learn to understand image context and generate missing pixels.

❓ Predict Output

advanced

What is the output shape after inpainting with this PyTorch model?

Given the following PyTorch code snippet for an image inpainting model, what is the shape of the output tensor?

Computer Vision

import torch
import torch.nn as nn

class SimpleInpaint(nn.Module):
    def __init__(self):
        super().__init__()
        self.encoder = nn.Sequential(
            nn.Conv2d(3, 64, 3, padding=1),
            nn.ReLU(),
            nn.MaxPool2d(2)
        )
        self.decoder = nn.Sequential(
            nn.ConvTranspose2d(64, 3, 2, stride=2),
            nn.Sigmoid()
        )
    def forward(self, x):
        x = self.encoder(x)
        x = self.decoder(x)
        return x

model = SimpleInpaint()
input_tensor = torch.randn(1, 3, 128, 128)
output = model(input_tensor)
output.shape

A torch.Size([1, 3, 128, 128])

B torch.Size([1, 64, 64, 64])

C torch.Size([1, 3, 64, 64])

D torch.Size([1, 64, 128, 128])

💡 Hint

Check how MaxPool2d and ConvTranspose2d change spatial dimensions.

MaxPool2d with kernel 2 halves the height and width from 128 to 64. ConvTranspose2d with kernel 2 and stride 2 doubles it back to 128. The output channels are 3, matching input channels.

❓ Metrics

advanced

Which metric best measures the quality of image inpainting results?

You want to evaluate how well your inpainting model fills missing parts of images. Which metric is most suitable?

A F1 Score

B Peak Signal-to-Noise Ratio (PSNR)

C Mean Squared Error (MSE) on class labels

D Accuracy

💡 Hint

Think about metrics that compare images pixel-by-pixel.

PSNR measures the similarity between the original and inpainted images, indicating reconstruction quality. Accuracy and F1 are for classification, not image similarity.

🔧 Debug

expert

Why does this image inpainting model produce blurry results?

You trained a CNN-based image inpainting model, but the filled areas look blurry and lack details. What is the most likely cause?

A The input images were too large, causing memory overflow.

B The model uses too many convolutional layers, causing overfitting.

C The model was trained using only pixel-wise loss like Mean Squared Error, which encourages averaging and smooth outputs.

D The model uses batch normalization, which reduces sharpness.

💡 Hint

Think about how loss functions affect output sharpness.

Pixel-wise losses like MSE penalize differences equally, leading the model to produce average pixel values in uncertain areas, causing blur. Using adversarial or perceptual losses can improve sharpness.