Custom trained model support

[junho2000]

Hi thank you for sharing your great work!

i am just curious that can i run executables with my custom trained model? my model has 12 class, not like coco dataset.

i export my best(yolov8n).pt model to onnx and then i run the detect_object_video then output is x,y, width, height are all 0 but only there is class name.

except for this everything works well!

[eeuwevandijk]

Which version of OpenCV (C++) do you use? I had several issues with Yolo C++ and ended up with a specific version of OpenCV: v4.7.0 (OpenCV v4.8+ did not work on my system: Jetson Orin AGX)

[junho2000]

I am working on my laptop 4060 mobile ubuntu20.04, i installed OpenCV 4.8.0

[cyrusbehr]

Please send the model @eeuwevandijk

[junho2000]

https://drive.google.com/file/d/13fV2zgIZAsq6ubgGY1CbQld6RmQIe3s_/view?usp=sharing this is my model(best.pt)

[cyrusbehr]

@eeuwevandijk OpenCV is not the issue here, I used version 4.8 and it worked fine for me with this model.

[cyrusbehr]

@junho2000

Your model does not have 12 classes like you mention, it has 9 classes:

That being said, my code does not read the classes automatically from the model. You need to manually specify them. You have two options.

1) Specify the classes using the --class-names command line argument. So it would look like this: ./detect_object_image --model ./best.onnx --input ~/Desktop/sign.png --class-names crosswalk-sign highway-enterance-sign highway-exit-sign no-entry-road-sign one-way-road-sign parking-sign priority-sign round-about-sign stop-sign 2) Modify the class names in the code. You need to make the change here. Replace what's there with your 9 classes.

Once you make the change, the model works fine. Here is a sample output with your model:

[junho2000]

@junho2000

Your model does not have 12 classes like you mention, it has 9 classes:

That being said, my code does not read the classes automatically from the model. You need to manually specify them. You have two options.

1) Specify the classes using the --class-names command line argument. So it would look like this: ./detect_object_image --model ./best.onnx --input ~/Desktop/sign.png --class-names crosswalk-sign highway-enterance-sign highway-exit-sign no-entry-road-sign one-way-road-sign parking-sign priority-sign round-about-sign stop-sign

2) Modify the class names in the code. You need to make the change here. Replace what's there with your 9 classes.

Once you make the change, the model works fine. Here is a sample output with your model:

Thank you so much..!

[junho2000]

Hi i tried both method (specify class name at CLI and revise yolov8.h class name) but still nothing detected, same happen

have you did extra thing to work fine?

[cyrusbehr]

What platform are you using? If you are running on a jetson, please check out the jetson branch.

[junho2000]

i am using on my laptop with geforce 4060 mobile