search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.72k stars 7.96k forks source link

How to import darknet into the micro-controller? For example: ARM cortex, Raspberry #1675

Open leekwunfung817 opened 6 years ago

leekwunfung817 commented 6 years ago

By the way, have anyone install the NVIDIA Geforce Graphic Card into the micro-controller?

PeterQuinn925 commented 6 years ago

There's a GPU on the Raspberry pi and there are some people trying to use it with Darknet. Other than that, your best bet is to use one of the USB compute sticks, like movidius with the Raspi. It would be really difficult to use a PC card with a Raspi. I've been experimenting with a USB smart camera called Jevois that you might look into.

AlexeyAB commented 6 years ago

@PeterQuinn925 Did you try to check speedup of detection https://github.com/AlexeyAB/darknet/blob/master/cfg/tiny-yolo_xnor.cfg on the Jevois?

(it should be trained by yourself)

PeterQuinn925 commented 6 years ago

Not yet. Jevois is using the NNPACK version and my first attempt to put your fork on it crashes the device.

leekwunfung817 commented 6 years ago

@AlexeyAB Yes, I need good performance and stable to long run darknet.

@PeterQuinn925 I cannot found the method to improve the performance in micro-controller by using Geforce Graphic Card.

I have been asked to produce the firmware in the IP camera, as I know the darknet are using pure C language, therefore darknet can burn in the firmware with graphic card driver.

By the way, well-known micro-controller may not support CUDA and CUDNN. Have anyone know about it?

That is quite a bit lot of trouble.

I need some solution. Thanks everyone.