Image dataset from folders in PyTorch - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of loading images with ImageFolder

What will be the output shape of the first image tensor loaded using torchvision.datasets.ImageFolder with default ToTensor transform?

PyTorch

from torchvision import datasets, transforms
transform = transforms.ToTensor()
dataset = datasets.ImageFolder(root='data/train', transform=transform)
image, label = dataset[0]
print(image.shape)

A torch.Size([H, W]) where H and W are image height and width

B torch.Size([H, W, 3]) where H and W are image height and width

C torch.Size([1, H, W]) where H and W are image height and width

D torch.Size([3, H, W]) where H and W are image height and width

💡 Hint

Remember that ToTensor converts images to CxHxW format.

The ToTensor() transform converts PIL images to tensors with shape (channels, height, width). For color images, channels = 3 (RGB).

❓ data_output

intermediate

Number of classes in ImageFolder dataset

Given an ImageFolder dataset loaded from a folder with subfolders 'cats', 'dogs', and 'rabbits', what is the value of dataset.classes?

PyTorch

from torchvision import datasets
 dataset = datasets.ImageFolder(root='data/train')
print(dataset.classes)

A ['rabbits', 'dogs', 'cats']

B ['dogs', 'cats', 'rabbits']

C ['cats', 'dogs', 'rabbits']

D ['cats', 'rabbits', 'dogs']

💡 Hint

ImageFolder sorts folder names alphabetically to assign classes.

ImageFolder sorts the folder names alphabetically to create the classes list. So the order is ['cats', 'dogs', 'rabbits'].

🔧 Debug

advanced

Identify error loading images with ImageFolder

What error will occur when running this code if the root folder contains a file that is not an image or folder? from torchvision import datasets dataset = datasets.ImageFolder(root='data/train') image, label = dataset[0]

PyTorch

from torchvision import datasets
dataset = datasets.ImageFolder(root='data/train')
image, label = dataset[0]

A PIL.UnidentifiedImageError

B IndexError

C FileNotFoundError

D TypeError

💡 Hint

ImageFolder tries to open all files as images.

If a file is not a valid image, PIL will raise UnidentifiedImageError when ImageFolder tries to open it.

🚀 Application

advanced

Transform pipeline for ImageFolder dataset

Which transform pipeline correctly resizes images to 128x128, converts to tensor, and normalizes with mean=0.5 and std=0.5 for all channels?

A transforms.Compose([transforms.ToTensor(), transforms.Resize((128,128)), transforms.Normalize((0.5,), (0.5,))])

B transforms.Compose([transforms.Resize((128,128)), transforms.ToTensor(), transforms.Normalize((0.5,0.5,0.5), (0.5,0.5,0.5))])

C transforms.Compose([transforms.Resize(128), transforms.ToTensor(), transforms.Normalize((0.5,0.5,0.5), (0.5,0.5,0.5))])

D transforms.Compose([transforms.Resize((128,128)), transforms.Normalize((0.5,0.5,0.5), (0.5,0.5,0.5)), transforms.ToTensor()])

💡 Hint

Resize should come before ToTensor, and Normalize after ToTensor.

The correct order is Resize, then ToTensor, then Normalize. Resize needs a tuple for size. Normalize expects 3 values for RGB channels.

🧠 Conceptual

expert

Why use ImageFolder for image datasets?

What is the main advantage of using torchvision.datasets.ImageFolder to load image datasets organized in folders?

A It automatically assigns labels based on folder names and loads images efficiently.

B It converts images to grayscale automatically.

C It only works with PNG images.

D It requires images to be in a single folder without subfolders.

💡 Hint

Think about how labels are assigned in ImageFolder.

ImageFolder reads images from subfolders and assigns labels based on folder names, simplifying dataset creation.