Anyway to increase performance?

Hello,

I tried the xnor SDK on the RPiZero (python) using the pet-human detector. I was able to get around 0.5 fps. Is there anyway to improve the performance? 

Michael Carlon

6replies Oldest first
  • Oldest first
  • Newest first
  • Active threads
  • Popular
  • Hi  Michael Carlon. We have models not yet on the Developer SDK that perform 7+ FPS on the Pi Zero. It may not yet be available until later releases. Could you share more on your use case and what you're ultimately trying to achieve with a fast model?

    Reply Like
  • Carlo C del Mundo Hi,

    I am working on a pet detector + alert system, using a raspberry pi 3 or 0. I am able to get acceptable performance on the rpi3 using tensorflow lite but was hoping xnor had a solution to get 3+ fps on a rpi0.

    Any chance I can try out the 7fps model you speak of?

    (the system needs to be fast enough to reliably detect quick pet-movements)

    Reply Like
  • We can potentially offer an evaluation of these faster models for the Pi 0 as part of a larger and broader subscription or license. For this, I recommend e-mailing sales@xnor.ai for next action steps.

    Thanks for your interest, Michael. If there's any feedback you'd like to give to us w.r.t. the SDK, we'd be happy to hear it as well.

    Reply Like
    • Carlo C del Mundo RE: SDK. The instructions for rpi0 python mention the .2 release (and paths...) Hence the provided instructions do not work cleanly out of the box.

      Also, it is not immediately obvious what path you should be in as the instructions jump around when read top to bottom.

      Re: faster model

      I'll contact sales and see if we can work something out 

      Reply Like
    • Michael Carlon Thank you for the feedback! Forwarded it to our SDK team.

      Reply Like 1
  • Hi Michael, I'd like to hear more about your SDK feedback. You can contact me directly via email todd@xnor.ai

    Reply Like
Like Follow
  • 7 mths agoLast active
  • 6Replies
  • 107Views
  • 3 Following