Instance Segmentation Export to YOLOV8-Seg Normalization Issue

Hi Everyone

I have a private project on Roboflow which I have successfully trained a Roboflow Model for Instance Segmentation.

However I would like to train a YOLOV8-Segmentation model on it on my machine.

When I export the dataset in YOLOv8 Oriented Bounding Boxes format like so:

I try to train then model, I get errors indicating that the dataset is not normalized.
/train/labels… 18110 images, 20 backgrounds, 6261 corrupt: 100%
ignoring corrupt image/label: non-normalized or out of bounds coordinates [ 1.0355]

I see the issue is there are some labels, with coordinates higher than 1.

I can look into writing a script to try and normalize the data or convert from V5, however it would really help if there was either a simple fix or if anyone can offer some other suggestions.

Hi @EthanShirk - what notebook or code are you using to train the model?

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