Recommendations to use hailo for long hours

I am developing an application with Raspberry Pi 5, 2 cameras and a hailo AI kit running yolov8pose model. My application needs to run continuously for 10-12 hours in a day. Since i am working with 2 cameras my overall cpu usage is always more than 50-60% at 10fps. Althogh I am working only with 10fps each to reduce the cpu post process load. So, I believe I am not utilizing too much hailo core.

So, do you have something to recommend for longer use, better performance, and, if possible, increasing fps a little bit? I have not done tests running continuously yet. I want to understand if there are any ideal usage recommendations.

Hi @saurabh,

The recommendations would depend on the models you are using, which ones are they?

@nina-vilela
I am using 2 parallel pipeline with 2 hailonet element. each running yolov8_pose model and post processing steps.

It would great if you could share a general way to decide good parametrs for using long hours on raspberrypi5…
What are some ideal parameters i need to comsider to make reliable… for ex fps, model size etc.
Do i need to think about it much?

I believe there is no direct way to measure usage of hailo8l. Only for hailo8.

I have more products to build using hailo8l, so I’m thinking in general terms.

Hi @saurabh,

For long hours you need to ensure to have the heatsink installed, no need to think about it besides that.

Answering your question about better performance and increasing FPS, try using the performance mode when compiling the model and using a higher batch size during inference.