Failure to Detection objects in the engine model(yolov9)

[jackkal2]

Hi, I recently came across your wonderful repository and tested it on the yolov9-s-converted network and found that the engine model performs worse than the pt model and the confidence value becomes very small when the box gets bigger. What do you think is the problem and is there a solution to fix this?

Also, can I convert the pt model to onnx first and then convert it to engine?

• Device: jetson orin nano
• OS: jetpack 5.1.5
• Cuda: 11.4
• TensorRT: 8.5

[jackkal2]

Also another problem is that the yolov9-s model does not detection any objects.
I also tested the original yolov9 repository model and this problem still existed.
Model conversion command:
sudo ./yolov9 -s ./yolov9-s.wts ./yolov9-s.engine s

Please guide me.

[mpj1234]

Hi, I recently came across your wonderful repository and tested it on the yolov9-s-converted network and found that the engine model performs worse than the pt model and the confidence value becomes very small when the box gets bigger. What do you think is the problem and is there a solution to fix this?

Also, can I convert the pt model to onnx first and then convert it to engine?

• Device: jetson orin nano
• OS: jetpack 5.1.5
• Cuda: 11.4
• TensorRT: 8.5

The engine generated by onnx cannot be used in this project.