Skip to content

Conversation

@KushajveerSingh
Copy link
Contributor

@KushajveerSingh KushajveerSingh commented Jul 8, 2020

Added rank_diff, top1_diff, top5_diff columns in the results of ImagenetV2 Matched Frequency, ImageNet-Sketch, ImageNet-Adversarial.

  • +26.1 (for positive difference)
  • -3 (for negative difference)
  • = (for no change)

Added a script results/generate_csv_results.py, which can be used for updating the csv files in the future. Add dummy values in the {rank, top1, top5}_diff columns when making an entry for a new model.

To add a new result.csv file in the future, only the results dictionary in the script needs to be updated. (This assumes that rank_diff, top1_diff, top5_diff columns are already present in the csv file. I did not include the if checks for that in the pull request, as it needs to be only done once).

@rwightman
Copy link
Collaborator

@KushajveerSingh awesome thanks, something I've been meaning to do for some time. Just a minor tweak request on the decimal digits. Could you limit the topx and crop pct entries to 3 decimal digits and the param count to 2?

@KushajveerSingh
Copy link
Contributor Author

Made the changes. Set precision of top1, top1_err, top1_diff, top5, top5_diff, top5_err, cropt_pct to 3 and param_count to 2.

Also, removed the TODO item from README.

@rwightman rwightman merged commit 0d5550c into huggingface:master Jul 9, 2020
@KushajveerSingh KushajveerSingh deleted the results_diff branch July 9, 2020 18:02
guoriyue pushed a commit to guoriyue/pytorch-image-models that referenced this pull request May 24, 2024
Add rank difference, and top-1/top-5 difference from ImageNet-1k validation for the 3 additional test sets
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants