Which Khadas SBC do you use?

Khadas VIM3

Which system do you use? Android, Ubuntu, OOWOW or others?

Ubuntu 20.04

Hi @Frank
I ran the yolov5 model as per the instructions from here, but there is little difference in runtime in CPU and NPU. (0.02 - 0.03 seconds)
I don’t understand why like that.
How can I run faster on NPU using dnn, is that possible?

@gonulsinem Have you tried running the demo provided on our documentation, how fast is it? When OpenCV DNN is running, it will try to run the convertible operators on the NPU, and the non-convertible operators will run on the CPU.

Yes, I tried. likewise, there is very little (0.02 - 0.03 seconds) difference between cpu and npu in the demo in the document.
When you say “convertible operators” do you mean “.nb” and “.so” files?

@gonulsinem No, opencv doesn’t use this is not a conversion method.We will ask the opencv side to issue a document or description.

1 Like

so it’s not possible for me to run it faster on NPU?

@gonulsinem DNN is officially released by OpenCV, we will try to make them optimize the model of the yolo series

Thanks for your answer, keep up the good work!