Just installed...

[wpiman]

Just installed Frigate in a Docker container in parallel with Blue Iris running on a Window VM. I bought a mini-pc and dropped a Coral TPU into one of the Slots and put a Intel Arc on one of the Oculink ports and a Nvidia 3060 on the other.

Interesting comparison with Blue Iris. I think they each have their own strengths. The plate detection for ALPR is seemingly better on BI with Codeproject. The built in alert with Frigate work pretty damn well.

With the new AMD CPUs I am just sipping CPU cycles.

 

[Flintstone61]

 

[Vettester]

I use Frigate to capture vehicle speed. It works fairly well.



I have it setup to capture anything over 25 mph which is the posted speed limit and I sort by estimated speed.

 

[GeoffS]

I just upgraded my frigate installation from a Raspberry Pi 5 to an ASUS NUC 14

Pi 5 had a HailoL8 TPU and got inference times of around 11 msec. The ASUS NUC 14 uses a Core 7 155H and using the GPU, I get inferences times of 6 msec with virtually no CPU usage.
I guess the Hailo can go in the drawer with the unsupported Coral TPU

 

[Mike]

I started messing around with Frigate a few months ago, have it running on an OptiPlex micro i5-12500T using facial recognition and speed detection which feeds my Home Assistant.

 

[wpiman]

I installed a Hailo 8 in addition to my Coral TPU. I disable the Coral TPU and will be trying just the Hailo for a day.

Claude Code is handling all of the config and docker compose stuff.

With the speed, it asks you how large the box is in meters. How do you figure that out?

 

[wpiman]

I just upgraded my frigate installation from a Raspberry Pi 5 to an ASUS NUC 14

Pi 5 had a HailoL8 TPU and got inference times of around 11 msec. The ASUS NUC 14 uses a Core 7 155H and using the GPU, I get inferences times of 6 msec with virtually no CPU usage.
I guess the Hailo can go in the drawer with the unsupported Coral TPU

Did you take a look and see how much of that GPU is being used? You might be using a lot of power and a Hailo uses a few watts.

I have an Intel Arc A380 and I am doing HW decoding on my camera and that is 15% of the card. My calculations (Claude) said if I do detection with the GPU I would max out the card.