Fix YOLOS image processor resizing

[qubvel]

What does this PR do?

I observed that the YOLOS model failed with OOM while training. After some investigation, I found that square images are not properly resized by YOLOS image processor even if do_resize=True. The size parameter is not respected even if the max size of the image is larger than longest_edge.

Example to reproduce:

import torch
from transformers import YolosImageProcessor

images = torch.randint(0, 256, (1, 3, 1500, 1500), dtype=torch.uint8)
processor = YolosImageProcessor(size={"longest_edge": 1333, "shortest_edge": 800}, do_resize=True)

# Resize the image
result = processor(images=images, return_tensors="pt")
print(result["pixel_values"].shape)

# >>> torch.Size([1, 3, 1488, 1488])

I am going to

1. submit a test that fails on square images
2. submit a fix

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.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[qubvel]

@amyeroberts it's ready 🙂

[amyeroberts]

Beautiful - thanks for fixing and for updating the tests!

[HindIbraheem]

--