Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[orientation] augment angle while training #1567

Merged
merged 3 commits into from
Apr 26, 2024

Conversation

felixdittrich92
Copy link
Contributor

@felixdittrich92 felixdittrich92 commented Apr 26, 2024

This PR:

  • augment randomly the angle while training (+-25 degree) (expand if not directly 0, 90, 180 or -90)
  • the idea: the current orientation models works pretty well if the real angle of the page or crop is close to [0, 90, 180, -90] but has some lack with larger deviations like >+- 5 degree

@odulcy-mindee We should test it on the TF orientation model training runs i think this will improve the robustness a lot and then do the same again with the PT models (or we test it directly with the PT models)

@felixdittrich92 felixdittrich92 added type: enhancement Improvement ext: references Related to references folder topic: character classification Related to the task of character classification labels Apr 26, 2024
@felixdittrich92 felixdittrich92 self-assigned this Apr 26, 2024
@felixdittrich92 felixdittrich92 marked this pull request as ready for review April 26, 2024 06:09
Copy link

codecov bot commented Apr 26, 2024

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 95.88%. Comparing base (630d925) to head (8b93ef4).

Additional details and impacted files
@@           Coverage Diff           @@
##             main    #1567   +/-   ##
=======================================
  Coverage   95.88%   95.88%           
=======================================
  Files         163      163           
  Lines        7649     7649           
=======================================
  Hits         7334     7334           
  Misses        315      315           
Flag Coverage Δ
unittests 95.88% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Copy link
Collaborator

@odulcy-mindee odulcy-mindee left a comment

Choose a reason for hiding this comment

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

Will the expand arg not mess with the input size of the image ? Or will it be correctly resized before entering the training pipeline ?

@felixT2K
Copy link
Contributor

Will the expand arg not mess with the input size of the image ? Or will it be correctly resized before entering the training pipeline ?

It's correctly resized before :)
Screenshot from 2024-04-26 15-38-49
Screenshot from 2024-04-26 15-38-15

@odulcy-mindee odulcy-mindee merged commit 5568612 into mindee:main Apr 26, 2024
75 of 81 checks passed
@felixdittrich92 felixdittrich92 deleted the augment-orient branch April 26, 2024 14:22
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
ext: references Related to references folder topic: character classification Related to the task of character classification type: enhancement Improvement
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants