Compose transforms in PyTorch - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is the purpose of Compose in PyTorch transforms?

Compose lets you combine multiple image transformations into one sequence. It applies each transform one after another to the input data.

beginner

How do you use Compose to apply a resize and then convert an image to a tensor?

transform = Compose([Resize((128, 128)), ToTensor()])

This creates a transform that first resizes images to 128x128 pixels, then converts them to tensors.

intermediate

What happens if you change the order of transforms inside Compose?

The transforms are applied in the order you list them. Changing order can change the final output. For example, Normalize() before ToTensor() will cause an error.

intermediate

Can Compose include custom transforms?

<p>Yes! You can write your own transform as a class or function and include it inside <code>Compose</code> along with built-in transforms.</p>

beginner

What is the output of applying Compose([Resize((64,64)), ToTensor()]) to an image?

The output is a PyTorch tensor representing the image resized to 64x64 pixels, ready for model input.

What does Compose do in PyTorch transforms?

A Loads images from disk

B Creates a new dataset

C Trains a neural network

D Chains multiple transforms to apply in sequence

Compose chains multiple transforms so they run one after another on the input.

Which of these is a correct way to use Compose?

A Compose([Resize((128,128)), ToTensor()])

B Compose([ToTensor(), Resize((128,128))])

C Compose(Resize((128,128)), ToTensor())

D Compose(ToTensor())

Transforms must be inside a list passed to Compose. Also, resizing should come before converting to tensor.

What type of object does ToTensor() produce?

A NumPy array

B PyTorch tensor

C Python list

D PIL Image

ToTensor() converts images to PyTorch tensors.

If you want to add a custom transform to Compose, what must it do?

A Train a model

B Load data from disk

C Return a transformed image or tensor

D Print the image

Custom transforms must take input and return the transformed output to work in Compose.

What happens if you put Normalize() before ToTensor() in Compose?

A Normalize will fail because input is a PIL image, not a tensor

B It works fine

C The image will be normalized twice

D The image will convert back to tensor

Normalize() expects a PyTorch tensor, so putting it before ToTensor() causes an error.

Explain how Compose helps when preparing images for a PyTorch model.

Think about how you get images ready step-by-step.

Describe what could go wrong if you change the order of transforms inside Compose.

Consider what happens if you convert to tensor too early.