My poor #Jetson #nano was no longer holding up while using #deepstack. I can't say why but it was regularly timing out and hanging.

So I pulled the trigger and installed #CodeProjectAI. However, CPAI is a .Net project and I think it's overhead is too much for the nano (Nevermind their build instructions being seriously wanting).

So I snagged a #Coral #TPU. As a USB device, this should be cake. I passed the port through to my Windows vm under #proxmox 8 (my need requires Windows, not *nix) loaded the Coral module and was getting 100ms responses. Then suddenly it stopped working. The USB device was no longer listed in device manager.

-What the crap-

Reboot it, pass it through again and it works for a handful of queries then the device disappears. I tried this by passing the port itself and the device ID, same experience both ways.

I have pcie devices passed through without issue but this silly thing will not stay attached. The host never loses it and it is visible constantly under lsusb, but the VM constantly "loses" it, so I can't rely on it. Back to #cpu processing I go.

I see similar bugs with some frigate users (, but they're largely all LXC (https://github.com/blakeblackshear/frigate/issues/1807). Is this a Windows issue? Proxmox? The TPU?

Ugh.

Why can't anything be simple and 'just work'?
#sysadmin #homelab

Coral USB on Proxmox LXC · Issue #1807 · blakeblackshear/frigate

Hello, Can't get Frigate working with Google Coral USB accelerator. I am running Proxmox 7.0, inside I have LXC container running Debian 10 turnkey core 16 image. Inside Debian I am running Frigate...

GitHub

Finally got #deepstack & #BlueIris configured with custom models on a #JetsonNano.

Processing times are down in the high 100ms to mid-200ms depending on image size.

It's not the 10ms I'd love, but since I don't have GPU offload in Proxmox (GPU gets used for other VM's and it's old so I can't split it for sharing among multiple hosts), it's as good as I'll get for a while.

In typical BI fashion it was more complex than needed. But between tcpdump and curl I figured out what it was doing and got it pointed in the right direction.

Now my doorbell gets rung when someone enters my patio or a car pulls into the driveway. Woo!

RT @[email protected]

With over 3.4 millions Docker installs as at today, I am glad to announce today that #DeepStack AI Server is now open source and the entire code base now available on GitHub.
Visit the linked below to read more about it.

https://medium.com/deepquestai/deepstack-ai-server-now-open-source-a0a35e53bd51

🐦🔗: https://twitter.com/OlafenwaMoses/status/1337035047371747330
DeepStack AI Server now Open Source

The leading AI Engine on the edge now open source on GitHub

Playing around with Deepstack AI. Image analysis to make my CCTV system smart and to get alerts when there is really a person and not clouds hovering over the sunlight.

Promising technology, seems to work pretty good!

#AI #CCTV #BlueIris #Deepstack #cp

https://deepstack.cc/

DeepStack AI Server

State-of-the-art AI APIs Offline and on the Cloud