Fix test_pipelines_video_classification that was always failing

[CalOmnie]

What does this PR do?

tests/test_pipelines_video_classification.py was always failing because the output of the video classifier, when run with one file path (as opposed to a list), is returning a list[dict] whereas the testing code was expecting a list[list[dict]], this PR corrects that assumption.

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.
It would be nice if someone could check that pytest tests/test_pipelines_video_classification.py actually fails for them too, it is possible that the issue comes from my environment

[Rocketknight1]

cc @ydshieh, can you confirm this test was failing?

[ydshieh]

@CalOmnie I am not able to see test_pipelines_video_classification.py failing on our CI report.

Could you provide a link to a job run page where this test is failing or a full command that run a specific test that shows this failure?

[CalOmnie]

Hey @ydshieh thanks for taking the time to go over this PR. I shared this command in the PR description but I probably should have made a dedicated section for it, I ran the tests using:

pytest tests/test_pipelines_video_classification.py

I assume it is not part of your automated testing, or it would have been found earlier. That or an issue with my own environment.

[ydshieh]

OK, it should be tests/pipelines/test_pipelines_video_classification.py. But with that, I can't see it from the Slack report as the message is truncated.

This test is indeed failing on main and fixed in this PR.

[ydshieh]

I find the docstring for VideoClassificationPipeline.__call__ in src/transformers/pipelines/video_classification.py is misleading:

        Return:
            A dictionary or a list of dictionaries containing result. If the input is a single video, will return a
            dictionary, if the input is a list of several videos, will return a list of dictionaries corresponding to
            the videos.

Even with a single video (path string), the returned value is still a list of dictionary instead of the mentioned dictionary in the above docstring. (It's a list of length top_k).

@CalOmnie Would you like to double check this part and update the docstring if necessary?

@Rocketknight1 The changes in this PR LGTM.

[CalOmnie]

@ydshieh Apologies for the wrong path. I updated the docstring to reflect that the pipeline now returns top_k results, even with a single video.

[ydshieh]

@Rocketknight1 Maybe you can double check the change in src/transformers/pipelines/video_classification.py and then we are ready to go?

[Rocketknight1]

@ydshieh the changes in the actual pipeline are just docstring edits, and I think the new docstring is more correct, so I'm happy to approve this!