Dataset bias in vision in Computer Vision - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to load images from a folder using torchvision.

Computer Vision

from torchvision import datasets, transforms

transform = transforms.Compose([transforms.Resize((128, 128)), transforms.ToTensor()])
dataset = datasets.ImageFolder(root='data/train', transform=[1])

Drag options to blanks, or click blank then click option'

Atransform

Btransforms

Cdatasets

DImageFolder

Attempts:

3 left

💡 Hint

Common Mistakes

Passing the module name 'transforms' instead of the transform pipeline.

Passing the dataset class 'datasets' instead of the transform.

Passing the class 'ImageFolder' instead of the transform.

The transform argument expects a transform pipeline, so 'transform' is the correct choice.

2fill in blank

medium

Complete the code to split the dataset into training and validation sets.

Computer Vision

from torch.utils.data import random_split

train_size = int(0.8 * len(dataset))
val_size = len(dataset) - train_size
train_dataset, val_dataset = random_split(dataset, [[1], val_size])

Drag options to blanks, or click blank then click option'

Aval_size

Btrain_size

Clen(dataset)

D0.8

Attempts:

3 left

💡 Hint

Common Mistakes

Swapping train_size and val_size in the split sizes.

Using the dataset length directly instead of the split sizes.

The first argument in the list should be the training size, which is 'train_size'.

3fill in blank

hard

Fix the error in the code that creates a DataLoader for the validation set.

Computer Vision

from torch.utils.data import DataLoader

val_loader = DataLoader(val_dataset, batch_size=32, shuffle=[1])

Drag options to blanks, or click blank then click option'

ATrue

BNone

CFalse

D0

Attempts:

3 left

💡 Hint

Common Mistakes

Setting shuffle to True for validation data.

Passing None or 0 which are invalid for shuffle parameter.

For validation data, shuffling should be False to keep order consistent.

4fill in blank

hard

Fill both blanks to create a dictionary comprehension that filters images with labels less than 5 and maps them to their file paths.

Computer Vision

filtered_images = {img_path: label for img_path, label in dataset.imgs if label [1] 5 and img_path.endswith([2])}

Drag options to blanks, or click blank then click option'

A<

B>

C'.jpg'

D'.png'

Attempts:

3 left

💡 Hint

Common Mistakes

Using '>' instead of '<' for label filtering.

Using '.png' when images are '.jpg'.

We want labels less than 5 and images ending with '.jpg'.

5fill in blank

hard

Fill all three blanks to define a function that checks if a dataset is biased by comparing class counts and returns True if any class count is less than 10.

Computer Vision

def is_biased(dataset):
    counts = {}
    for _, label in dataset:
        counts[label] = counts.get([1], 0) + 1
    return any(count < [2] for count in counts.values()) or len(counts) < [3]

Drag options to blanks, or click blank then click option'

Alabel

B10

C5

Dlen(dataset)

Attempts:

3 left

💡 Hint

Common Mistakes

Using wrong key in counts.get().

Using wrong threshold values for counts or class number.

We count labels by key 'label', check if any count is less than 10, and check if number of classes is less than 5.