Deepstack-gpu realtime processing

Hi I’m running deepstack and home assistant on virtual machines with GPU, and I’m trying to process camera images in real time with No luck, I can fire process manually but not found a way to do so automatically in real time as anyone else achieve this also? Would appreciate a little bit of help all the points in the right direction.
thanks in advance Andy.

@robmarkcole is very knowledgable in this area. Look out his threads, or with luck he’ll pop in and point you in the right direction.

1 Like

The current integration is not really designed for real time, you can do 1 FPS in reality. You should checkout Frigate for true real-time processing

Thanks for the recommendation will look into that hopefully it’s not too complicated for me to get somewhere with it, I think I will continue messing around with deep stack because it’s up and running at the moment.Thanks.

I wrote a service (.net core over docker) that samples snapshots of camera and process them against the API of deepstack,
It takes around 70 msec to get the image (from BlueIris) and process takes around 100 msec, so in general I can process 5 images in a second,
I’m running it to sample 2 camera sources every 1 second and it works amazingly…

https://hub.docker.com/r/eladbar/ip2mqtt

2 Likes

Hi sorry to bother you again but is it possible that you could send the GitHub file or make it public thanks.
Andy.

@Andy_Nunes, I’m doing realtime processing of my front IPC stream to open a wicket by face recognition. I’m getting around 15 fps in object detection (coral) and 3 fps in face recognition (deepstack on Jetson). I wrote a standalone python program that does the processing and communicates with HA via MQTT. Happy to share if interested.

1 Like

@sickidolderivative yes please share

It’s fast & dirty and specific to my home setup, but perhaps someone finds parts of it useful:

20+ FPS object detection with Google Coral AI is accessible through fork of @robmarkcole’s excellent tensorflow rest server:

1 Like