Image-to-image transformation in Prompt Engineering / GenAI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the main goal of image-to-image transformation models?

Choose the best description of what image-to-image transformation models do.

A Classify images into categories based on their content.

B Detect objects and draw bounding boxes on images.

C Generate images from random noise without any input image.

D Convert an input image into another image with desired changes, like style or content.

💡 Hint

Think about models that change one image into another.

Image-to-image transformation models take an input image and produce a new image that has been changed in some way, such as changing style, color, or adding effects.

❓ Model Choice

intermediate

Which model architecture is commonly used for image-to-image transformation tasks?

Select the model architecture best suited for image-to-image transformation.

A Convolutional Autoencoder with skip connections (U-Net)

B Recurrent Neural Network (RNN)

C Transformer for text generation

D Fully connected feedforward network

💡 Hint

Look for a model that preserves spatial details and can reconstruct images.

U-Net is a convolutional autoencoder with skip connections that helps preserve image details, making it ideal for image-to-image tasks.

❓ Metrics

advanced

Which metric best measures the quality of generated images in image-to-image transformation?

Choose the metric that evaluates how close the generated image is to the target image in terms of pixel-level similarity.

A Mean Squared Error (MSE)

B Accuracy

C Perplexity

D BLEU score

💡 Hint

Think about a metric that calculates average squared differences between pixels.

Mean Squared Error measures the average squared difference between pixels of generated and target images, indicating similarity.

🔧 Debug

advanced

What error will this image-to-image transformation training code raise?

Consider this Python snippet for training a model. What error occurs when running it? ```python import torch from torch import nn model = nn.Sequential( nn.Conv2d(3, 64, 3, padding=1), nn.ReLU(), nn.Conv2d(64, 3, 3, padding=1) ) input_image = torch.randn(1, 3, 256, 256) output = model(input_image) loss_fn = nn.MSELoss() # Target image has wrong shape target_image = torch.randn(1, 3, 128, 128) loss = loss_fn(output, target_image) ```

A TypeError: loss_fn() missing 1 required positional argument

B SyntaxError: invalid syntax in model definition

C RuntimeError: The size of tensor a (256) must match the size of tensor b (128) at non-singleton dimension 2

D No error, code runs successfully

💡 Hint

Check if output and target images have the same shape before computing loss.

The output image shape is (1,3,256,256) but target image shape is (1,3,128,128). MSELoss requires both inputs to have the same shape, causing a RuntimeError.

❓ Hyperparameter

expert

Which hyperparameter adjustment is most likely to improve image sharpness in a GAN-based image-to-image model?

You notice generated images are blurry. Which change is most effective to improve sharpness?

A Decrease the learning rate of the generator

B Increase the discriminator capacity or depth

C Reduce the batch size drastically

D Remove batch normalization layers from the generator

💡 Hint

Sharper images often come from a stronger discriminator that pushes the generator harder.

Increasing discriminator capacity helps it better distinguish real from fake images, forcing the generator to produce sharper images.