Skip to content

Conversation

pmeier
Copy link
Collaborator

@pmeier pmeier commented Feb 5, 2021

Fixes #3304.

Copy link
Contributor

@datumbox datumbox left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks!

@pmeier
Copy link
Collaborator Author

pmeier commented Feb 5, 2021

Unfortunately it is hard to add a testcase for this, since it requires to have at least 256 images in the mock up. Let me know if we want this anyway.

@datumbox
Copy link
Contributor

datumbox commented Feb 5, 2021

Yes adding a test on the near future would be nice, but I would not want to block the PR waiting for that because it resolves a bug. Perhaps loading a tiny dataset with few records and then repeat them post-loading to create a bigger one? I'm sure you will find something. :)

@datumbox datumbox merged commit ced96a0 into pytorch:master Feb 5, 2021
@pmeier pmeier deleted the fix-stl10-folds branch February 5, 2021 11:36
facebook-github-bot pushed a commit that referenced this pull request Feb 10, 2021
Summary:
* fix STL10 folds

* use np.int64 over np.long

Reviewed By: fmassa

Differential Revision: D26341419

fbshipit-source-id: 5f59667c4e53c4e375665b3609c09ed72f3a0207
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

torchvision.dataset.STL10.__load_folds() loads image indices using uint8.

3 participants