Hi @nihui ,

Thank you for your efforts.

I tried Yolov5 NCNN model (using c++ https://github.com/Tencent/ncnn/blob/master/examples/yolov5.cpp ) and found the inference is slow compared to python model. On further analysis found that the output extraction step takes majority of the inference time.

ex.extract("output", out); - takes an average of 72.5666 millisecond out of total Inference time is 98.093 millisecond. Please advise what steps can be done to reduce the time. Thanks.