How to solve an error in custom OBB model detection?

General Discussion
2

Hi @Xiangsheng_Wang,

Hi, this error usually means the model was compiled with a single merged output instead of the separate convolution end nodes that the postprocessor expects. You might want to check how you exported your ONNX - for YOLOv8-OBB you typically need 9 separate end nodes (6 for standard detection + 3 for the angle branch), and you may need a custom .alls script to define them properly during compilation.

This might help: Is there any oriented bounding box model object detection

Thanks,
Michael.