[Config Support]: Configuration is very suboptimal, and heavily relies on CPU #19783
Replies: 4 comments 3 replies
-
One of your biggest causes of false positives comes from the fact that you are using the default model that ships with Frigate. This model is based on the COCO dataset and is not trained specifically on security camera images. There ways to mitigate false positives through object filters, but you could consider using a more accurate model with your Nvidia card like YOLOv9. You could also try a Frigate+ model. See the official documentation on the ONNX detector and models for your Nvidia card: https://docs.frigate.video/configuration/object_detectors/#onnx |
Beta Was this translation helpful? Give feedback.
-
@inkeep can also make some other suggestions on improving your config, including optimizing restreaming, not using |
Beta Was this translation helpful? Give feedback.
-
Here are some targeted suggestions based on the official documentation:
For more detailed tuning, see: 📚 Sources: Have another question? Just tag @inkeep. |
Beta Was this translation helpful? Give feedback.
-
But what about the CPU usage? I can see a potential to get rid of CPU detection when i migrate to ONNX, but go2rtc or even recording seem to be taking a lot of CPU time in my setup, basically clogging up a very powerful server |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Describe the problem you are having
Hiya. The hardware im running Frigate on is R720 poweredge with 2x Xeon, 260GB of RAM and Nvidia RTX 5000.

Over the years, my Frigate config survived many different additions and changes of hardware, many sessions of tinkering with it, and to be honest, i no longer even remember why i have put some of the things in the past. Anyhow, it works very suboptimally. It still uses CPU heavily, and the cameras can be temperamental (usually everything works tip top after reboot, but after a few days, it can lose main streams for some of the cameras, things arent getting recorded, etc). On the top of that, the object detection is quite bad - gives a lot of false positives. Around Frigate 0.14.0 i have tried TensorRT with YOLO model and in terms of false positibes it was even worse. Also i cannot force WebRTC for the Reolink doorbell to work.
I would appreciate if someone with deeper understanding of Frigate could review it and advice how to improve/optimise it.
Doorbell is Reolink Doorbell 2K PoE, and then the majority of the cameras are Dahuas (different models) configured similarly to this one:
(exceptions are baby room and kitchen cameras)
Version
0.16.0
Frigate config file
Relevant Frigate log output
Relevant go2rtc log output
Frigate stats
Operating system
Debian
Install method
Docker Compose
docker-compose file or Docker CLI command
Object Detector
Coral
Screenshots of the Frigate UI's System metrics pages
Any other information that may be helpful
No response
Beta Was this translation helpful? Give feedback.
All reactions