FOMO Inference and Web Stream ESP32

Help
#1

Question/Issue:
Is it possible/ worth attempting to performance FOMO inference and stream the output on a ESP32 at the same time?

Project ID:
182955

Context/Use case:
I am just getting started with microcontrollers and recently purchased both an ESP32-CAM(AI Thinker) and ESP-EYE. I have exported a FOMO model as an Arduino library and confirmed that it runs on both devices. I have also tested both devices as camera webservers using GitHub - easytarget/esp32-cam-webserver: Expanded version of the Espressif ESP webcam
I can’t find evidence of anyone attempting to combine these functionality on one device, I don’t see any reason why it wouldn’t be possible besides memory issues, but I imagine the performance may be very poor. Is there any precedent for this? Should I keep the devices as webcams and leave inference to a more powerful device?

#2

Hello @seangaff,

The inference will be indeed a slow but it should work.
You probably won’t be able to achieve more than 1 with both FOMO and a webserver running (with a small image).
And you’ll probably need to optimize the memory consumption too.

Best,

Louis

#3

I know this is an old post and I know this looks like awful self-promotion, but I did this because me too couldn’t find any existing example on the web. Yes it is possible, yes it is complicated, no it isn’t THAT slow, only runs on ESP32S3 boards. https://twitter.com/EloquentArduino/status/1722956050280656904