Hi,
thank you for this really cool tool! :)
I have noticed a discrepancy of the image dimensions when exporting from custom vision:
-
A downloaded image is of real size: 1280 x 960
-
But in the json file the dimensions are wrong:
"images": [
{
"id": 1,
"width": 1200,
"height": 900,
"file_name": "images/000000000001.jpg",
...
-
The bounding boxes in the annotations are then also relative to these wrong dimensions... which of course leads to incorrect tagging downstream.
(I am using the workaround to read real images again to normalize the wrong dimensions and annotations. )
Thank you and best regards,
Robert