Data loading with torchvision in Computer Vision - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is the purpose of torchvision.datasets in data loading?

It provides ready-to-use datasets for computer vision tasks, making it easy to download, load, and preprocess common image datasets.

beginner

Explain the role of torch.utils.data.DataLoader.

It wraps a dataset and provides an iterable over the data with support for batching, shuffling, and parallel loading using multiple workers.

beginner

How do transforms help in data loading with torchvision?

Transforms apply preprocessing steps like resizing, cropping, normalization, and data augmentation to images before feeding them to the model.

beginner

What does setting shuffle=True in DataLoader do?

It randomizes the order of data samples each epoch to help the model generalize better by preventing learning the order of data.

intermediate

Why use num_workers in DataLoader?

It allows loading data in parallel using multiple subprocesses, speeding up data preparation especially for large datasets.

Which torchvision class is used to load datasets like CIFAR10 or MNIST?

A torch.nn.Module

B torchvision.transforms

C torch.utils.data.DataLoader

D torchvision.datasets

torchvision.datasets provides classes to download and load popular vision datasets.

What does the batch_size parameter in DataLoader control?

A Number of samples per batch

B Number of workers loading data

C Whether to shuffle data

D Image size after transform

batch_size sets how many samples are returned in each batch during iteration.

Which transform would you use to convert images to tensors?

A transforms.ToTensor()

B transforms.Normalize()

C transforms.Resize()

D transforms.RandomCrop()

ToTensor converts PIL images or numpy arrays to PyTorch tensors.

Why is shuffling data important during training?

A To increase batch size

B To prevent model from memorizing data order

C To reduce image size

D To speed up loading

Shuffling helps the model learn general patterns rather than the order of data.

What is the effect of increasing num_workers in DataLoader?

A More data shuffling

B Larger batch size

C Faster data loading by parallelism

D Transforms applied multiple times

More workers load data in parallel, reducing waiting time for batches.

Describe the steps to load and prepare an image dataset using torchvision.

Think about dataset, transforms, and DataLoader roles.

Explain why data augmentation is important and how torchvision supports it during data loading.

Consider how transforms help the model see different versions of images.