Problems in replicating the performance

[vvigilante]

Hello!
I'm trying to replicate your experiment: Raspberry Pi 3, Darknet19, NNPACK=1,ARM_NEON=1,NNPACK_FAST=1, all other switches are off (GPU, CUDNN, OPENCV, OPENMP, DEBUG, QPU_GEMM)

I should get: 1.3 (first frame), 0.66 (subsequent frames)
But I get: 6 seconds for the first frame and 3 for the subsequent ones, that is 4.5 times slower.

Do you think I'm missing some step in configuration? Did I set the wrong switches? Do I need to modify something in particular in NNPACK's init.c?

Thank you!

[shizukachan]

I think I derped. On a Pi3 @ 900MHz:
From 2017-12-26 commit I got 6.44/3.33 on Darknet19 (224x224)
From 2018-05-31 commit I got 7.02/3.88 on Darknet19 (256x256)

I get 2.10/1.05 on Darknet (256x256)... oops, I mislabeled the benchmark results!
The update to the yolov3 tree also updated all of the cfg files, so the benchmark results (other than yolov3) are all likely out of date as it looks like they are all 256x256 (and I tested on 224x224).