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

fix: linknet hyperparameters postprocessing + demo for rotation model #865

Merged
merged 3 commits into from
Mar 22, 2022

Conversation

charlesmindee
Copy link
Collaborator

This PR adjusts the Linknet parameters in the postprocessing to have better performances, and includes the linknet_resnet18_rotation model in the streamlit demo. Pages are rendered only for straight models (for now), and we don't write the words above each boxes if the model is a rotation model.
This demo will be improved later, but for the moment it allows anyone to quickly see the performances of our rotation model without modifying the behavior for all other models.

Any feedback is welcome!

This is how it renders for the demo:
image

@charlesmindee charlesmindee added type: enhancement Improvement module: models Related to doctr.models ext: demo Related to demo folder topic: text detection Related to the task of text detection labels Mar 21, 2022
@charlesmindee charlesmindee requested a review from fg-mindee March 21, 2022 08:35
@charlesmindee charlesmindee self-assigned this Mar 21, 2022
@codecov
Copy link

codecov bot commented Mar 21, 2022

Codecov Report

Merging #865 (bbdcce3) into main (fc8efc9) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##             main     #865   +/-   ##
=======================================
  Coverage   94.84%   94.84%           
=======================================
  Files         133      133           
  Lines        5200     5200           
=======================================
  Hits         4932     4932           
  Misses        268      268           
Flag Coverage Δ
unittests 94.84% <100.00%> (ø)

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

Impacted Files Coverage Δ
doctr/models/detection/linknet/base.py 88.69% <100.00%> (-0.10%) ⬇️
doctr/utils/visualization.py 89.65% <100.00%> (+0.08%) ⬆️
doctr/models/zoo.py 100.00% <0.00%> (ø)
doctr/utils/data.py 88.67% <0.00%> (ø)
doctr/datasets/svt.py 97.22% <0.00%> (ø)
doctr/datasets/cord.py 97.43% <0.00%> (ø)
doctr/datasets/ic03.py 97.36% <0.00%> (ø)
doctr/datasets/ic13.py 96.15% <0.00%> (ø)
doctr/datasets/svhn.py 94.73% <0.00%> (ø)
doctr/models/_utils.py 97.59% <0.00%> (ø)
... and 39 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update fc8efc9...bbdcce3. Read the comment docs.

@fg-mindee
Copy link
Contributor

Thanks @charlesmindee 🙏
What about the page synthesis? Is it still working?

Copy link
Contributor

@fg-mindee fg-mindee left a comment

Choose a reason for hiding this comment

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

Thanks for the PR! I think there are a few things that might need tweaking for the next release, but this can be done in a later PR 👍

demo/app.py Show resolved Hide resolved
doctr/models/detection/linknet/base.py Show resolved Hide resolved
doctr/models/detection/linknet/base.py Show resolved Hide resolved
doctr/utils/visualization.py Show resolved Hide resolved
@fg-mindee fg-mindee added this to the 0.5.1 milestone Mar 21, 2022
@charlesmindee charlesmindee merged commit 9d03085 into main Mar 22, 2022
@charlesmindee charlesmindee deleted the fix_linknet_param branch March 22, 2022 09:49
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
ext: demo Related to demo folder module: models Related to doctr.models topic: text detection Related to the task of text detection type: enhancement Improvement
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants