I've run yolov2 on my Raspberry Pi 3 but I got 1 frame each 13s (which is pretty bad for live object detection). I've also tried an SSD but the results were terrible too. Is there anything I could do to make it faster?

-I've tryied this library for YOLO https://github.com/allanzelener/YAD2K and a keras implementation of SSD

-My camera is a simple Microsoft webcam medium quality

  • 1
    Welcome to the site. It's impossible to know whether your approach could be made faster if we can't see how you did it - can you edit your question (don't use comments) to include the exact command string you're using with yolov2, and how you might have any peripherals (cameras etc.) attached?
    – goobering
    Commented Feb 19, 2017 at 12:50
  • A sure way to increase your processing speed is to decrease the resolution of the image. Commented May 18, 2017 at 12:00
  • Can you please describe the setps that you used. Did u used Raspberian OS? Thanks
    – victorp
    Commented Jul 21, 2017 at 4:16

3 Answers 3


OK after a long search I 've found a reliable version of yolo called YAD2K. I have use tiny-yolo which takes about 1.5s per frame on a raspberry pi 3.

  • what FPS did u mange to achieve ? @simonepi Commented Sep 15, 2017 at 13:24
  • @ArsenalFanatic 1.5s/frame = 0.67FPS Commented Apr 14, 2018 at 14:07

I managed to had yolo v3 tiny run on my Pi 3 model B+ at 1FPS. I compiled the darknet source code with NNPACK option, which significantly improved the FPS (from ~0.1FPS to 1FPS). Check by blogs for results http://funofdiy.blogspot.com/2018/08/deep-learning-with-raspberry-pi-real.html

  • 1
    The native darknet performs pretty bad on CPU. That's why you need NNPACK, which optimizes neural network performance on multi-core CPU. I think Pi 3 Cortex-A53 has four cores so using NNPACK you will be expecting to see 3~4x acceleration. Still, Yolo2 is big and will be slow on RPI. Try yolo tiny version
    – Xiang Zhai
    Commented Aug 30, 2018 at 15:45

Running YOLO on the raspberry pi 3 was slow. What i did was use Intel's Movidius NCS it was a little tricky getting it all setup, but that was mainly due to the fact it had just came out and had a few bugs. Got it to work using Stretch OS on the Pi 3. Then was able to run it on the Pi zero. The speed you get with it is wicked quick. https://www.pyimagesearch.com/2018/02/19/real-time-object-detection-on-the-raspberry-pi-with-the-movidius-ncs/

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.