Best detector for use with Double-Take and Frigate on a Raspberry Pi?

[jerome71]

Hi.
I have (successfully I think 😊) set up Frigate on a Raspberry Pi 5 8 GB + Coral USB TPU. 8 cameras, including 3 on a different location, integrated thanks to Tailscale. Some H264 re-encoding (CPU only on the Pi 5) that is responsible for most of the CPU usage.
It works great, and much better than I expected from such a setup.
The Pi still had 50% CPU power available.

I have now set up Double Take, with DeepStack. On the same Raspberry Pi. I get some results - they are not great (which I expected because the cameras are usually far from the subjects, so they appear in quite low resolution).

My question is: what would be your recommendation be for a detector for face recognition, supported by Double Take, that works on a Raspberry Pi?
My understanding is that Compreface is X64 only, so does not work on the Pi's ARM64.
So should I try CodeProject.AI Server? Or Facebox?
Or should I stick with Deepstack?

Bonus question: is there a way I can use the Coral USB TPU in both Frigate and the Double Take detector?
I tried using the Coral USB TPU simultaneously in 2 Frigate instances and could not make it work.

Thanks in advance! 😊🙏

[molnarti]

hey, what did you end up with? i am in the same situation, both all detectors (deepstack, facebox or compreface) seem to be dead with last releases 2 years ago...

also, how did you integrate your offsite cameras? where are you running the tailscale nodes on them?

[elektrinis]

just confirming that these detectors (deepstack, facebox or compreface) are dead and actually really awful, unusable. They detect my wife as me etc.
It looks like open source face detection is dead for now. On frigate/HA at least.

[NickM-27]

I disagree, using compreface it worked really well for me. Compreface worked with multiple different models and many users made the mistake of not customizing which model they were using.

And also, Frigate 0.16 already has face recognition and ALPR implemented.