transformers.image_transforms.normalize wrong types

[CalOmnie]

What does this PR do?

transformers.image_transforms.normalize types std and mean as Iterable, and checks for that, but calls len on them. The proper type for a sized iterable is Sequence, this PR adds that.

Fixes #35772

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[Rocketknight1]

Yes, this is correct. Thanks for the fix!

[tomasruizt]

This PR was probably supposed to be a refactor, and should not change the execution paths of the code. However, it turns out that numpy arrays are Iterables but not Sequences, and therefore the execution paths changed. Client code that used to pass numpy arrays for mean and std fails now. We found this because it broke our MiniCPM video-chat pipeline. Case here: https://huggingface.co/openbmb/MiniCPM-V-2_6/discussions/53

from typing import Iterable, Sequence
import numpy as np

x = np.ones(3)
print(isinstance(x, Iterable))
#> True
print(isinstance(x, Sequence))
#> False

I can see that the type hint does not include numpy arrays, but passing a numpy array (which I guess is the intention of the type hint Sequence[float] / Iterable[float]) is reasonable, too. Would you say that client code is responsible for adjusting? Or could we revert? Or what would guys you suggest?

Best :)

Edit: I found that some client libs like vLLM are adjusting accordingly vllm-project/vllm@e8616d7

[CalOmnie]

Oh, you're right, I went one level too high in the typing, the type needed was Collection not Sequence, see here

I will have a PR out tonight unless someone beats me to it.

>>> from typing import Collection
>>> import numpy as np
>>> a = np.ones(10)
>>> isinstance(a, Collection)
True

[CalOmnie]

Alright I created a fix and modified some tests to make sure it worked here