When I try to export my annotations for a yolov8-pose project, I get many images with annotations that are negative or out-of-bounds. What can I do about this?
- Project Type: Pose Detection
- Operating System & Browser: Edge - Windows
Hey, how is it going?
Can you share a little bit more about your dataset? Generally, what is the size of your images that are getting the wrong annotation exports?
Hello,
I did a bit of experimentation, and it seems like these messages disappear when I remove the following preprocess steps:
- Auto-Orient: Applied
- Resize: Fill (with center crop) in 640x640
When these preprocessing steps are applied, roboflow shows some of the larger images in the dataset get cropped in a way that makes the annotations appear out of the image.
Is there a way to apply these preprocessing steps without this happening?
This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.