Misslabeling bounding boxes

Community Help
1

I’m having an issue when exporting the dataset. I originally created the labels on Labelimg, a Python library, then I imported them to the roboflow environment to correct some labels. Then my problems start. When I export the labels, they appear to be misplaced both in roboflow and in my function call in Python. I followed some steps that I saw here in the forum, like deleting the old versions of the dataset; now the roboflow images seem to be ok, but every time I export, I see that in Python some are misplaced.

  • Project Type:Object detection*
  • Operating System & Browser:Windows 11, Google Chrome*
  • Project Universe Link or Workspace/Project ID:weathering-stages*

102507
1025071218×647 47.6 KB

2

Hi @wintermute, are you able to provide some of the files that you’re importing to make it easier to reproduce the issue you’re having?

Also, can you explain how you are exporting labels and checking them for correctness?

Cheers!
Tony

3

Hi @tonylampada I’m exporting with reoriented on auto and resized to 1024x1024 too, but I managed to solve the problem by exporting in the original size. When I exported to Python, I used PIL and xmletree to work with these files. The problem with roboflow, i think, is the auto-flip that is made on the images, and besides that, when the image is resized, I think the information in the xml is not updated too along with the box dimensions, so this causes a problem with shape correspondence in Python too. But thanks for your response.

4

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.