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

Meeting 27.03.2024 #1

Open
3 of 7 tasks
derrickmuheki opened this issue Mar 27, 2024 · 7 comments
Open
3 of 7 tasks

Meeting 27.03.2024 #1

derrickmuheki opened this issue Mar 27, 2024 · 7 comments

Comments

@derrickmuheki
Copy link
Collaborator

derrickmuheki commented Mar 27, 2024

  • Documentation
  • Workflow illustration
  • Template matching @khufkens @derrickmuheki
  • Jungle Weather info routine porting
  • Fred's Imagemagick: Textdeskew, innercrop, @derrickmuheki
  • Unit tests - Git hub actions
  • Translate to python packages
@khufkens
Copy link
Collaborator

Added dockerfile routine in #2

khufkens added a commit that referenced this issue Mar 27, 2024
@khufkens
Copy link
Collaborator

ImageMagick doesn't seem to perform well.

Initial start of a python based approach is started here (stackexchange demo - needs testing):
https://github.com/bluegreen-labs/Data_Rescue_Congo_DRC/blob/main/src/crop_border.py

@khufkens
Copy link
Collaborator

khufkens commented Mar 27, 2024

@derrickmuheki Illustration of the workflow is made using http://draw.io, both the original xml and a png are included as of #3

@khufkens
Copy link
Collaborator

ported over the JR info on the template routine as of #4

@derrickmuheki
Copy link
Collaborator Author

@khufkens been trying out template matching further today (see added module and guides txt file from GIMP). While it led to nice results for some images, in others it was way off (see attached examples) especially where there table isn't detected as the largest contour perhaps due to the paper quality or edges detected as largest contour. I will try to figure this out tomorrow. Also I was thinking whether there might be a way to relate the bounding boxes gotten as a result of template guides (hor and vert) as shown in the new template_matching module, with the bounding boxes gotten from contours (earlier done in the table_detection_model.py). Any advice will be greatly appreciated.
bounding_boxes_trial_1.pdf
bounding_boxes_trial_2_with_error.pdf

@khufkens
Copy link
Collaborator

khufkens commented Apr 3, 2024

@derrickmuheki could you provide a link to the code (I think you haven't pushed things yet).

If I look at the error image I see a lot of noise, maybe tweaking some of the binarization parameters might help. My setup assumes even lighting.

The CLAHE filter might mess things up, as it amplifies contrast of blobs and spots unduly.
https://github.com/bluegreen-labs/jungleweather/blob/80a6ffaf2cc57d71123097692f8366c0b8f22515/src/match_template.py#L188

@derrickmuheki
Copy link
Collaborator Author

I have now pushed the changes. My bad, thought i had done so yesterday. Going to try out this CLAHE Filter too

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

No branches or pull requests

2 participants