Yolo11 Boundary box is always rectangle

roboflow / notebooks

Examples and tutorials on using SOTA computer vision models and techniques. Learn everything from old-school ResNet, through YOLO and object-detection transformers like DETR, to the latest models like Grounding DINO and SAM.

https://roboflow.com/models

5.6k stars 882 forks source link

Yolo11 Boundary box is always rectangle #325

Open darthvaddor opened 1 month ago

darthvaddor commented 1 month ago

I am trying to finetune yolo on my dataset to predict the object, but even if the object is skewed the boundary box is always and not a skewed rectangle. While training I labelled my dataset carefully, not just using standard rectangles, rather skewed 4 sided polygons to outline the object. But why is it that while predicting it doesnt do this

darthvaddor commented 1 month ago

@SkalskiP @capjamesg @artyaltanzaya @gasparitiago

LinasKo commented 1 month ago

Hi @JAVARSHA 👋

There are 3 types of annotations a model can support

Bounding boxes, which are always implied to be rectangles without any rotation.
Masks, which operate on pixel-level and can take any shape.
Oriented Bounding Boxes (OBB) which are rectangles, but rotated. This is the least popular method.

Very precisely detecting a skewed rectangle can only be done with masks and instance segmentation. I would suggest trying out oriented bounding boxes as well.

Hope this helps!