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

How do I correct line recognition? #33

Open
zabak opened this issue May 13, 2020 · 6 comments
Open

How do I correct line recognition? #33

zabak opened this issue May 13, 2020 · 6 comments
Assignees
Labels
enhancement New feature or request

Comments

@zabak
Copy link

zabak commented May 13, 2020

When correcting OCR for National Museum, the regions were mostly recognized without problems. When correcting OCR, however, I found cases where part of word was not detected as part of line. Sometimes, it has not been detected at all:
image
Other times: there were overlapping or duplicate detections:
image
How should I treat it when correcting?

Should I write the whole line, even the letters that are not part of the line as marked on the image? Should I write the same text twice, when correcting overlapping lines?

@zabak
Copy link
Author

zabak commented Mar 17, 2022

This is really a problem - please implement support for fixing the text line boundaries and use them to make a better model.

@zabak
Copy link
Author

zabak commented Mar 17, 2022

@michal-hradis
Copy link
Collaborator

We are dubugging layout editor. Normally, you edit text in /ocr/show_results/... The new interface is running at /ocr/show_results_new/... Just manually change the URL. I'll probably have to explain how the interface works.

@zabak
Copy link
Author

zabak commented Mar 20, 2022

Yes please, I will need some explanation, the interface is not intuitive. Default zoom should zoom to fill the window width or height, not zoom out so much as it does now. There is no way to delete a row. I have no idea how to select two rows to join them. How to resize a row, how edit a shape of a region.

@zabak
Copy link
Author

zabak commented Feb 15, 2023

@michal-hradis please add the explanation here. Also, how to revert OCR without losing the manually edited baselines.

@michal-hradis
Copy link
Collaborator

michal-hradis commented Feb 15, 2023

Text transcriptions can be generated again and again without any loss of manual text corrections. Text line detection can not be repeated without loosing manual corrections.

How to edit text lines:

  1. Select by left click.
  2. Press CTRL to show controll points.
  3. Drag controll points with mouse left button. You can delete controll points by moving the line end over them. You can't add controll points.
  4. Change line height by dragging "top" and "bottom" controll points et the end and beginnig of a line.

To delete lines :

  1. Select line.
  2. Press ALT-B or press the "Delete line" button.
  3. This option can be rolled back, but the line disapears only after the document page is reloaded.

Alternative way to delete lines which can not be rolled back and which tends to delete whole text region if you are not carefull at the moment:

  1. Right click on a line.
  2. Slect delete in the context menu, but check carefully that it shows: "Delete row" and not "Delete region" - the second one shows by mistake if you move the mouse when clicking.

Add lines:

  1. Select a region.
  2. Slect tool: "Create new row (baseline).
  3. Create a baseline by left clicking. Finish baseline by pressing enter or right click.
  4. Select line hight - top part and bottom part separately. Use left click to set the heights.

Edit regions:

  1. Select region.
  2. Press alt to show region controll points.
  3. Drag the controll points around.

You can further:

  • Add regions
  • Delete regions (deletes all associated text lines)
  • Merge text lines (does not work properly at the moment)

@michal-hradis michal-hradis added the enhancement New feature or request label Feb 15, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request
Projects
None yet
Development

No branches or pull requests

3 participants