[Examples] Computing FID in unconditional image generation

[alirezadizaji]

What does this PR do?

Since sometimes there's a need for quantitative assessment of image generation, this PR brings an example of FID computation, particularly for unconditional image generation. Using it is quite straightforward: simply by passing the --compute_fid argument in CLI, now it's possible to log the FID score in the tracker.

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?
• Did you read our philosophy doc (important for complex PRs)?
• 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.

[alirezadizaji]

@sayakpaul

[sayakpaul]

Thank you! Would it not make sense to default to clean-fid instead?

[sayakpaul]

I believe both of these could be clubbed together?

[alirezadizaji]

Thank you! Would it not make sense to default to clean-fid instead?

Yes, it's a great idea to include clean FID. We could also add an option for clean-fid to let the user decide. It could be implemented in a separate PR after this one.

[alirezadizaji]

@sayakpaul

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[sayakpaul]

@alirezadizaji gentle ping regarding the open comments.

[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.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[alirezadizaji]

Thank you! Would it not make sense to default to clean-fid instead?

One nice property of the current FID computation is that it only processes real images once before the beginning of training (lines 549 to 566), which is faster than passing them every time to compute the scores, as clean-fid does.

[sayakpaul]

I think multiple works now have reported enough issues about FID and have resorted to using lcean-fid despite the computational intensity. So, I think that would be preferrable.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.