There are other options. I use deepstack on my nas for object recognition so that I only get alerted by the camera if it recognises vehicles or people in our driveway

[–] NoRamyunForYou 3 points 7 months ago (1 children)

What type of hardware are you running if you don't mind me asking?

[–] eagleeyedtiger 3 points 7 months ago (1 children)

It's running on a 10 year old xeon. I already had Unraid running as my Nas that has a bunch of docker containers for Plex, Homeassistant, etc. Deepstack is running on there as a docker container.

FWIW, I setup my security camera 1-2 years ago, so there might be better options now. However there was also a big shortage of Coral TPU's at the time, even though I did consider using Frigate. I ended up settling on BlueIris for recording as I wasn't happy with what the free options offered back then. Unfortunately BlueIris is Windows only, so I picked up an ex-lease PC which runs headless just for BlueIris and storing the footage.

I've been thinking lately of updating the hardware in the Nas, as there is a huge performance jump if I move to one of the new Raptor/Alder lake Intel chips with quick sync, particularly for transcoding for Plex. But it's going to be at least a couple of grand to update everything properly, which I don't have to spend right now :(

[–] NoRamyunForYou 3 points 7 months ago (1 children)

During my research, I've been seeing a lot of talk around Docker. Is that sort of like in between a Venv & VM?

My understanding is that is you use a Coral TPU, it basically allows you to run things such as Image recognition on much lesser hardware than would otherwise be possible? In theory, does that mean you don't need the top of the line everything else if you're trying to run something like that?

[–] eagleeyedtiger 3 points 7 months ago (1 children)

I’m not going to pretend like I’m an expert at either of these. Unraid has come a long way now that installing app containers is pretty trivial. I am going to link this reddit comment that does a pretty good ELI5: Here

With the Coral, I think it’s literally the tensor processing unit (hence TPU) or coprocessor that is found usually on GPU’s for AI and machine learning. So you can use it with lower powered devices that can’t fit a discrete GPU or offload the processing to the TPU instead of using CPU/GPU. I think. Don’t quote me on that 😆

[–] NoRamyunForYou 3 points 7 months ago

Looking at what others have done online, it seems the corals can really benefit inference speeds for image recognition. Hopefully I'll be putting my order in to test it out :)