There seems to be a bug when opening the annotation file and visually adjust the size and location of the previously labeled bboxes？ about labelimg HOT 7 CLOSED

I have fixed the bug by modifying the Line 121 of pascal_voc_io.py as following:
points = [(xmin,ymin), (xmin,ymax), (xmax, ymax), (xmax, ymin)] -->
points = [(xmin, ymin), (xmax, ymin), (xmax, ymax), (xmin, ymax)]

Anyway, thanks a lot :)

from labelimg.

@tzutalin You can just load an image, annotate, save and then close the project. ( I did this with Open Dir and set the default annotation folder). Close it and re-open with Open Dir, then try to edit rectangles ... you'll see that it fails to keep the aspect ratio and messes the box.

from labelimg.

Hi @George-Zhu ,
Can you share me with the examples which have problems? Or do you think it is ok to use the below changes without side effect?
points = [(xmin,ymin), (xmin,ymax), (xmax, ymax), (xmax, ymin)] -->
points = [(xmin, ymin), (xmax, ymin), (xmax, ymax), (xmin, ymax)]

from labelimg.

@George-Zhu Thanks for the fix! Saved my day!
@tzutalin @George-Zhu : I wonder why is this caused during parsing the xml with wrong order of points, I guessed there was an issue with the type of shape or something, then I found this solution and wondered, can you explain the cause?

from labelimg.

HI @nkhdiscovery ,
If you modify the Line 121 of pascal_voc_io.py as following, that can fix your the issue, right?
points = [(xmin,ymin), (xmin,ymax), (xmax, ymax), (xmax, ymin)] -->
points = [(xmin, ymin), (xmax, ymin), (xmax, ymax), (xmin, ymax)]

from labelimg.

Hi @tzutalin ,
Yes it worked, I'm using it right now.
Great work by the way :)
I will spend a pull request if I improved anything, but not for this issue right now as I still have not figured the reason well.

Actually, I think this has no further harm as I had a quick review of the code, changing this line makes the order clock-wise and later in adding these points to shape and building the canvas (which is handled by Qt I guess), the resulting canvas would be obviously treated differently.
I let you know more if I figured it completely.

from labelimg.

@nkhdiscovery ,
Very appreciated ! I am looking forward to your PR!!!!

from labelimg.