MNN time consumption statistics about lffd-mnn HOT 7 CLOSED

Hello,@xjcvip007 , when the input size is 320x240, num_thread=4, the inference time on PC is about 33.4731 ms . If you want to know the mobile time cost, you can test it by yourself.

from lffd-mnn.

Thank for your reply, I test it on iphone 7p with default size (640*480) and default config, its inference time is almost 100ms, maybe a little slow.

from lffd-mnn.

@xjcvip007, hello ,which scale_num=5 or 8? ,Actually, you can set the input size 320*240 on your phone! the Inference time of LFFD with the input shape of 320x240 is about 20ms on the Qualcomm Snapdragon 632 CPU when I test it with ncnn. and mnn may be faster.

And , are you sure you have set the config.type =MNN_FORWARD_METAL; the num_thread=?

from lffd-mnn.

I use your default config with MNN_FORWARD_CPU and thread=2~

from lffd-mnn.

@SyGoing
I am getting 90 ms for 4 threads on i7-7th generation, scale is 5.
Quite slow actually.

from lffd-mnn.

@dexception ,hello you test it on iphone 7? scale_num is 5? I suggest you set the input_resized size 320x240, In that way ,the time maybe faster, and the precision is also high, I test it on android with that set, about 20ms

from lffd-mnn.

@SyGoing
I have tested it on PC and these are horrendous stats.

from lffd-mnn.