Yunet quantized model is slower

[MathieuLalaque]

Hello ! Thanks for the great work.

I have been using the Yunet model and tried the quantized version to speed up inference but I got slower results, both in my own code and trying your demo. I checked the benchmarks and the quantized model is announced to be slower too : is this an expected behavior ?

For context I use default backend and I have a Intel(R) Core(TM) i5-10300H CPU. To be clear, I loaded the int8 onnx file from the github and did not use the quantization script.

[fengyuentau]

is this an expected behavior ?

Yes for now. Optimization on the int8 inference on default backend is still in progress opencv/opencv#23689.