🐛 Describe the bug

Setting "hello" instead of "train", "valid", "test" or "all" to split argument of CelebA() gets the direct error message as shown below:

from torchvision.datasets import CelebA

my_data = CelebA(
    root="data",
    split="hallo" # Here
) # Error

ValueError: Unknown value 'hallo' for argument split. Valid values are {'train', 'valid', 'test', 'all'}.

But setting 3.14 or ["train"] to split argument of CelebA() gets the indirect error message as shown below:

from torchvision.datasets import CelebA

my_data = CelebA(
    root="data",
    split=3.14
) # Error

AttributeError: 'float' object has no attribute 'lower'

from torchvision.datasets import CelebA

my_data = CelebA(
    root="data",
    split=["train"] # Here
) # Error

AttributeError: 'list' object has no attribute 'lower'

So, they should return the direct error messages as shown below:

ValueError: Unknown value 3.14 for argument split. Valid values are {'train', 'valid', 'test', 'all'}.

ValueError: Unknown value ["train"] for argument split. Valid values are {'train', 'valid', 'test', 'all'}.

Versions

import torchvision

torchvision.__version__ # '0.20.1'