Wrong bounding boxes on export with augments (YOLOv8 - Oriented Bounding Boxes)

Project Type: Object Detection
OS: Ubuntu 22.04
Browser: Firefox

I’m having a problem when exporting my dataset with augments, generating inaccurate bounding boxes (it uses oriented bounding boxes).

Here’s, an example:

  1. How it looks on Roboflow

  2. it’s how it looks when i export it to Yolov8-OBB format, download it and draw the bounding box on the image

  3. For comparison, here’s the same the image , but i exported the dataset without augments:

Augments used:

  • Brightness: Between -15% and +15%
  • Exposure: Between -10% and +10%

All of them are using the Auto-Orient preprocessing, without it the boundings boxes are even more broken).

Is anyone else having this problem?