Augmentation policy search (AutoAugment) in Computer Vision - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to import the AutoAugment policy from torchvision.

Computer Vision

from torchvision.transforms import [1]

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

ANormalize

BAutoAugment

CRandomCrop

DToTensor

Attempts:

3 left

💡 Hint

Common Mistakes

Importing unrelated transforms like RandomCrop or Normalize instead of AutoAugment.

The AutoAugment class is imported from torchvision.transforms to apply augmentation policies automatically.

2fill in blank

medium

Complete the code to create an AutoAugment transform with the CIFAR10 policy.

Computer Vision

transform = AutoAugment(policy=[1])

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

AAutoAugmentPolicy.CIFAR10

BAutoAugmentPolicy.IMAGENET

CAutoAugmentPolicy.SVHN

DAutoAugmentPolicy.MNIST

Attempts:

3 left

💡 Hint

Common Mistakes

Using the ImageNet or SVHN policies which are not suitable for CIFAR10 images.

The CIFAR10 policy is used here to apply augmentation suitable for CIFAR10 images.

3fill in blank

hard

Fix the error in applying AutoAugment transform to the dataset.

Computer Vision

dataset = datasets.CIFAR10(root='./data', train=True, download=True, transform=[1])

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

AAutoAugment()

Btransform()

CAutoAugment

Dtransform

Attempts:

3 left

💡 Hint

Common Mistakes

Passing the class name without parentheses, causing a type error.

The transform argument expects a callable transform object, so AutoAugment() must be instantiated.

4fill in blank

hard

Fill both blanks to create a composed transform with AutoAugment and normalization.

Computer Vision

transform = transforms.Compose([
    AutoAugment(policy=[1]),
    transforms.Normalize(mean=[2], std=[0.247, 0.243, 0.261])
])

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

AAutoAugmentPolicy.CIFAR10

B[0.4914, 0.4822, 0.4465]

C[0.5, 0.5, 0.5]

DAutoAugmentPolicy.IMAGENET

Attempts:

3 left

💡 Hint

Common Mistakes

Using ImageNet policy or wrong mean values for normalization.

The CIFAR10 policy is used for AutoAugment, and the mean values correspond to CIFAR10 dataset normalization.

5fill in blank

hard

Fill all three blanks to define a training loop that applies AutoAugment and computes accuracy.

Computer Vision

for images, labels in dataloader:
    images = images.to(device)
    labels = labels.to(device)
    outputs = model([1])
    loss = criterion(outputs, labels)
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()
    _, predicted = outputs.[2](1)
    correct += (predicted == labels).[3]().item()

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

Aimages

Bmax

Csum

Dargmax

Attempts:

3 left

💡 Hint

Common Mistakes

Passing labels to the model instead of images.

Using argmax instead of max which returns both values and indices.

Using count() instead of sum() to count correct predictions.

The model input is images. The max function returns max values and indices, and sum counts correct predictions.