Deterministic order of classes and images for the Omniglot dataset

[V0XNIHILI]

PR for #8677

By sorting the output of the list_files() and list_dir() helper functions, the order of samples in the Omniglot dataset can be made deterministic across OSes which helps reproducibility across machines for few-shot learning tasks where the accuracy can change significantly when different classes are used at a certain instant.

[pytorch-bot[bot]]

:link: Helpful Links

:test_tube: See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/8678

• :page_facing_up: Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[facebook-github-bot]

Hi @V0XNIHILI!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at [email protected]. Thanks!

[NicolasHug]

Sorry @V0XNIHILI , if we were to sort the input we'd be running the risk of outputting the dataset in a different order than how it was originally packaged. Ensuring that the order is reproducible within an arch seems reasonable , but across OSs it may not be in scope

[V0XNIHILI]

I understand! What would be the best way to go about it then? For example, other torchvision datasets do not have this issue as they retrieve the filenames (and as such, also the order) from a separate file instead of listing files and dirs.

For example: would it make sense to store a copy of the original order in file and then load this?

That guarantees:

• Same order across OSes
• Same ordering as original dataset

Let me know, happy to work on this!

[NicolasHug]

I think storing a copy of the original order would have to be part of the original dataset, but that's not somethin torchvision can control, only the dataset authors can do something about that.

I'm not sure there's an easy way to enforce what you need unfortunately, sorry :/

BTW, according to SO this seems to be more of a file-system dependent issue rather than OS-dependent: https://stackoverflow.com/questions/31534583/is-os-listdir-deterministic