With `p=0`, setting a list of no transforms to `transforms` argument of `RandomApply()` doesn't get error

[hyperkai]

🐛 Describe the bug

With p=1, setting a list of no transforms to transforms argument of RandomApply() gets the error as shown below. *You can see this issue as well:

from torchvision.datasets import OxfordIIITPet
from torchvision.transforms.v2 import RandomApply

my_data = OxfordIIITPet(
    root="data",                   # ↓↓↓↓↓↓↓
    transform=RandomApply(transforms=[], p=1)
)

my_data[0][0] # Error

UnboundLocalError: cannot access local variable 'outputs' where it is not associated with a value

But with p=0, setting a list of no transforms to transforms argument doesn't get error as shown below:

from torchvision.datasets import OxfordIIITPet
from torchvision.transforms.v2 import RandomApply

my_data = OxfordIIITPet(
    root="data",                   # ↓↓↓↓↓↓↓
    transform=RandomApply(transforms=[], p=0)
)

my_data[0][0] # No error

Versions

import torchvision

torchvision.__version__ # '0.20.1'

[NicolasHug]

Thanks for the report. The current behavior sounds reasonably to me. p=0 really is an edge case that corresponds to a pass-through, so it's OK not to error.