Did I misunderstand data.DataSet.__getitem__()?

1

This is very weird. I have loaded CIFAR10 dataset:

import torchvision.datasets as datasets
import torchvision.transforms as transforms
dataloader = datasets.CIFAR10
transform_train = transforms.Compose([
    transforms.RandomCrop(32, padding=4),
    transforms.RandomHorizontalFlip(),
    transforms.ToTensor(),
    transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
trainset = dataloader(root='./data', train=True, download=True, transform=transform_train)

Then I find the image returned by trainset.getitem and the one accessed directly are different.

The getitem in class CIFAR10(data.Dataset) says

    def __getitem__(self, index):
        img, target = self.data[index], self.targets[index]

        # doing this so that it is consistent with all other datasets
        # to return a PIL Image
        img = Image.fromarray(img)

        if self.transform is not None:
            img = self.transform(img)

        if self.target_transform is not None:
            target = self.target_transform(target)

        return img, target

But if I implement getitem() directly and run this

by_get_item = trainset[0][0]
directly = trainset.transform(Image.fromarray(trainset.data[0]))
print(torch.sum(torch.abs(
    by_get_item - directly
)))

it prints a non-zero and the value keeps changing. Why does this happen?

2

Since you are using random transformations, you’ll get randomly transformed samples every time.
Remove RandomHorizontalFlip and RandomCrop and you should get the same outputs.

3 Likes
3

Thanks a lot. I’m not familiar with torchvision transforms