Using Roboflow to train dataset.
Using OAK-D camera.
Dataset trained using Yolo v5.
Annotation boxes are as large as the screen but reading what the object is correct.
(https://i.stack.imgur.com/fsLAY.jpg)
This is an example of one picture I used. For each pool ball/billiard ball I have over 20 images.
I’ve tried retraining in a different model.
I’ve also tweaked the confidence and overlap.
Nothing seems to work I’m sure I’m overlooking it any help would work.
