Fix: siglip image processor rgb_convert is not being applied correctly.

[jp1924]

What does this PR do?

        # All transformations expect numpy arrays.
        images = [to_numpy_array(image) for image in images]

        if do_convert_rgb:
            images = [convert_to_rgb(image) for image in images]

The convert_to_rgb function is not being applied because ndarray values are being input
(convert_to_rgb simply returns the input value if it's not a PIL.Image).

If an image with RGBA or similar format is input, the conversion doesn't work properly,
causing the following error in the infer_channel_dimension_format method:

Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/transformers/image_utils.py", line 254, in infer_channel_dimension_format
    raise ValueError("Unable to infer channel dimension format")
ValueError: Unable to infer channel dimension format

transformers version: 4.46.0.dev0

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?

@amyeroberts

[molbap]

Good catch @jp1924 ! - I checked, it seems Siglip was the only exception with the wrong order of operations.

[jp1924]

@molbap
Oh, thank you for reviewing.
Separately, I noticed that do_convert_rgb is not implemented in ViTImageProcessor. Was this intentional?

[molbap]

There's no particular reason why - I have no opposition to introducing this in another PR :)

[jp1924]

@molbap
Oh yeah? So I can just open the PR and add it?

[molbap]

@jp1924, sure, no opposition to it - the main thing we are careful about is not breaking backwards compatibility. In this case, no previous behaviour can be broken since you're adding a functionality, so feel free to open a PR :)

[qubvel]

Thanks for the fix, indeed convert_rgb is not supported for numpy arrays! I suppose we can merge this fix!

Also would be great to update convert_rgb to avoid silently skipping numpy arrays.

[ArthurZucker]

thanks 🤗