Custom transforms in PyTorch - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to define a custom transform class that converts an image to grayscale.

PyTorch

import torchvision.transforms as transforms

class ToGrayScale:
    def __call__(self, img):
        return img.[1]('L')

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

Ato_grayscale

Bconvert

Cgrayscale

Dconvert_grayscale

Attempts:

3 left

💡 Hint

Common Mistakes

Using a non-existent method like 'to_grayscale' causes errors.

Forgetting to specify the number of output channels.

The convert method is used to change the image mode, and 'L' mode means single channel grayscale.

2fill in blank

medium

Complete the code to apply the custom transform to an image.

PyTorch

from PIL import Image

img = Image.open('sample.jpg')
transform = ToGrayScale()
gray_img = transform.[1](img)

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

A__call__

B__call

Ctransform

Dapply

Attempts:

3 left

💡 Hint

Common Mistakes

Trying to call a non-existent method like 'apply'.

Using incorrect method names like '__call' without trailing underscores.

The __call__ method allows the instance to be called like a function. Using transform(img) calls __call__ internally.

3fill in blank

hard

Fix the error in the custom transform that should normalize a tensor image.

PyTorch

import torch

class NormalizeTensor:
    def __init__(self, mean, std):
        self.mean = mean
        self.std = std
    def __call__(self, tensor):
        return (tensor - self.mean) / [1]

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

Aself.mean

Bmean

Cstd

Dself.std

Attempts:

3 left

💡 Hint

Common Mistakes

Using 'mean' or 'self.mean' in the denominator causes wrong normalization.

Forgetting to use 'self.' prefix for instance variables.

Normalization subtracts the mean and divides by the standard deviation stored in self.std.

4fill in blank

hard

Fill both blanks to complete a custom transform that randomly flips an image horizontally with 50% chance.

PyTorch

import random
from PIL import Image

class RandomHorizontalFlip:
    def __call__(self, img):
        if random.random() [1] 0.5:
            return img.[2](Image.FLIP_LEFT_RIGHT)
        return img

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

A>

B<

Ctranspose

Dflip

Attempts:

3 left

💡 Hint

Common Mistakes

Using '<' instead of '>' changes the flip probability.

Using a non-existent method like 'flip()' causes errors.

The condition checks if a random number is greater than 0.5 to flip. The transpose() method with Image.FLIP_LEFT_RIGHT flips the image horizontally.

5fill in blank

hard

Fill all three blanks to create a custom transform that converts an image to tensor and normalizes it.

PyTorch

import torchvision.transforms.functional as F

class ToTensorNormalize:
    def __init__(self, mean, std):
        self.mean = mean
        self.std = std
    def __call__(self, img):
        tensor = F.[1](img)
        tensor = F.[2](tensor, self.mean, [3])
        return tensor

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

Ato_tensor

Bnormalize

Cself.std

Dto_tensor_normalize

Attempts:

3 left

💡 Hint

Common Mistakes

Using a non-existent function like 'to_tensor_normalize'.

Passing mean or std incorrectly without 'self.'.

to_tensor converts the image to a tensor. normalize applies normalization using mean and standard deviation. The standard deviation is passed as self.std.