Detection speedup for microcontroller usage

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )

http://pjreddie.com/darknet/

Other

21.77k stars 7.96k forks source link

Detection speedup for microcontroller usage #2644

Open AbanoubMamdouh opened 5 years ago

AbanoubMamdouh commented 5 years ago

Hello, I am trying to run tiny-YOLOv3 for a real time detection application on a CPU, as you know, most commercial micro-computers/controllers do not support CUDA and OpenCL. How to improve the detection speed on the CPU usage? I utilised openMP and it speeded the detection by x4, but still one image takes 2 seconds in order to be classified and localized. Which parameters should I change? The anchors? The number of layers? How? Any suggestions? Thanks in advance!

AlexeyAB commented 5 years ago

@AbanoubMamdouh Hi,

What CPU do you use? Is it ARM CPU?

AbanoubMamdouh commented 5 years ago

Yes it is, I am working on the ARTIK 710 module

EhsanVahab commented 5 years ago

@AbanoubMamdouh Have you ever seen the following package? https://github.com/shizukachan/darknet-nnpack