[Model] Add SLANeXt Model Support

[liu-jiaxuan]

What does this PR do?

Fixes # (issue)

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.

[zucchini-nlp]

Hey, thanks for the PR! Could you rebase and update a few moments to align with the latest API?

[zhang-prog]

@zucchini-nlp We’ve finished the changes. PTAL.

Discussion:
BaseImageProcessor.resize affects accuracy, because this model uses OpenCV-based resizing during both training and inference. To avoid directly relying on cv2.resize, we implemented a NumPy-based approximation of cv2.resize in the slow image processor to better align with the original accuracy.

However, NumPy-style operations do not seem appropriate for the fast image processor, so we are inclined to think that this model does not need a SLANeXtImageProcessorFast.

What do you think? Also, for the sake of code cleanliness and readability, would it be acceptable to use cv2.resize directly in the slow image processor?

[zucchini-nlp]

What do you think? Also, for the sake of code cleanliness and readability, would it be acceptable to use cv2.resize directly in the slow image processor?

Unfortunately we don't have cv2 in dependencies list, so for slow processor we need to stick to numpy. In Fast processor, we can use numpy and torchvision, does torchvision.F.resize work for you?

[zucchini-nlp]

I see a few comments not yet resolved, commented there. As for resize, in slow processor we can use PIL and I believe it will work for you case. PIL has all the same resize types as torchvision

Actually we can even delete the slow processor, as fast processor is enough and we're defaulting to load a fast processor anyway. Much easier for you than fixing the code prob

[vasqu]

@liu-jiaxuan @zhang-prog I pushed my fixes:

• We need to recheck the integration tests in general and the keys have changed because the base model prefix was wrong
• I didn't find any usage of self.loc_generator which I commented out and added todo, pls check

[vasqu]

run-slow: slanext

[github-actions]

Workflow Run ⚙️

This comment contains run-slow, running the specified jobs:

models: ["models/slanext"]
quantizations: []

[github-actions]

CI Results

Workflow Run ⚙️

Commit Info

Context	Commit	Description
RUN	8e53663d	workflow commit (merge commit)
PR	b04e6606	branch commit (from PR)
main	e8712550	base commit (on main)

Model CI Report

❌ 1 new failed tests from this PR 😭

• slanext:
  tests/models/slanext/test_modeling_slanext.py::SLANeXtModelIntegrationTest::test_inference_table_recognition_head (✅ ⟹ ❌)

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: auto, slanext

[vasqu]

Checked locally with our runners, everything passes now! Merging

Thanks a lot @liu-jiaxuan and @zhang-prog for sticking through

[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.