Setting a `float` type tensor to `RandomPosterize()` works against the error messages

[hyperkai]

🐛 Describe the bug

Setting a complex and bool type tensor to RandomPosterize() gets the error messages saying Number of value bits is only defined for integer dtypes as shown below:

from torchvision.transforms.v2 import RandomPosterize

rp = RandomPosterize(bits=1, p=1)
              # ↓↓↓↓↓↓↓↓↓↓
rp(torch.tensor([[5.+0.j]])) # Error

              # ↓↓↓↓↓↓↓↓
rp(torch.tensor([[True]])) # Error

TypeError: Number of value bits is only defined for integer dtypes, but got torch.complex64.

TypeError: Number of value bits is only defined for integer dtypes, but got torch.bool.

But setting a float type tensor to RandomPosterize() works against the error messages above as shown below:

from torchvision.transforms.v2 import RandomPosterize

rp = RandomPosterize(bits=1, p=1)
              # ↓↓↓↓↓↓
rp(torch.tensor([[5.]]))
# tensor([[0.5000]])

Versions

import torchvision

torchvision.__version__ # '0.20.1'

[NicolasHug]

Same as in other issues: the transforms can be a no-op randomly. This is expected.