Yolov8 -Non Normalized or Out of Bounds

🤝 Community Help
1

Project Type: Image recognition model
I am trying to build an image recognition model, and have used roboflow for creating bounding boxes. When I started trianig it gave me the following error:
train: WARNING :warning: /home/ubuntu/train/images/o19_jpg.rf.a7b50eca866a8b68db70771d7105fa83.jpg: ignoring corrupt image/label: non-normalized or out of bounds coordinates [ 1.084 2.0452 1.0811 2.0804]
train: WARNING :warning: /home/ubuntu/train/images/o9_jpg.rf.6a498ca246d33de99c0ced867e4b3dec.jpg: ignoring corrupt image/label: non-normalized or out of bounds coordinates [ 4.429 4.4762 4.4263]

This was for almost 2k pictures, my total dataset is 14k images. I have used normal bounding boxes for annotations, and have used the following augmentations:

Preprocessing

Auto-Orient: Applied

Filter Null: Require all images to contain annotations.

Augmentations

Outputs per training example: 2

90° Rotate: Clockwise, Counter-Clockwise, Upside Down

Shear: ±10° Horizontal, ±10° Vertical

I have tried looking at everything, and can not find the actual source of issue. Can someone please assist me in this. Also I thought it was a problem when creating a version. So I remade, and it started freezing at zipping files. Because of these I ended up making another version, but still the error persisted.

Assistance will be appreciated.

3

Hi @Hamza_Shariq - did you get this resolved?

I recommend creating another version and try training another model. It’s likely we’ve fixed the underlying issue since you first posted.