What frame rate can be achieved on Raspberry Pi 3B?

Help
#1

Hello
I have questions…
What frame rate can be achieved on Raspberry Pi 3 with Object Detection Model (FOMO)?
it works slowly for me and the frame is delayed by 1-2 seconds at a picture size of 96X96 RGB

Edge Impulse Studio predicts me
for float — 3 ms
for int8 – 108 ms
(does the model with float mean 36 times faster?
is it because of the FPU?)
and the model at work gives (Found bounding boxes [30 ms])
how to calculate frequency?
FPS=1/Inference time
what to insert instead of Inference time

How can I increase the frequency without changing the size of the picture?
Thanks

#2

Hi @Edesign,

We haven’t ran benchmarks on RPi 3B as we officially support RPi 4 only. The numbers showed in the UI are for the latter.
It seems surprising to have float running faster, I’ll check with our engineering on this. For the FPS, you’re correct about the formula.

You can increase the frequency by selecting the smaller MobileNet model (MobileNet v2 0.1).

Aurelien

#3

Thanks Aurel, I will try MobileNet v2 0.1.
I would be grateful if you check with the engineers