44ms inference speed on Frigate with i5-13500 and 2 Reolink cams?

[tough yarrow]

I read other people have sub 10ms with these new Intel CPUs?

[mint wagon]

Looks like is running on cpu not GPU

[tough yarrow]

I just upgraded from an ancient server (4790k) so looks like I need to go back trhfoguh the setup process again then

is this in config? or in the docker settings?

[mint wagon]

Itll most likely be a system setup issue, need to install correct driver

[tough yarrow]

okay thanks will go back through the setup guide

my config just incase its related: https://pastebin.com/CGtuyA7P

[tough yarrow]

guessing this is the culprit

should be iHD?

[mint wagon]

definitely

[tough yarrow]

so I changed that, still getting 44ms

this doesnt look right

15mhz lol

[mint wagon]

need to use QSV for hwaccel

and you need to setup openvino object detector

[tough yarrow]

have I not done that right?

https://pastebin.com/CGtuyA7P

[mint wagon]

qsv looks right

[tough yarrow]

so just add this: detectors:
cpu1:
type: openvino
device: CPU

now Frigate wont boot

god setting Frigate up makes me feel stupid lol

[mint wagon]

You need to change the model

Just copy what is in the docs

[tough yarrow]

where do I do that? is that in the config? or somewhere else

I'm not sure where it covers this in the docs

this is what I have in my confiog now:

mqtt:
enabled: false
ffmpeg:
hwaccel_args: preset-intel-qsv-h264
detectors:
cpu1:
type: openvino
device: CPU
go2rtc:

and it no l onger starts

@mint wagon could you sense check this please? https://pastebin.com/4Mb9s4fM

because Frigate boots, but now I get this:

I can click through and view the cameras

I give up lol somehow fked my docker up and now its orphaned no idea why

[mint wagon]

This config is not from the docs

You just delete the detectors config and use the documentation openvino config. Just use the copy button

https://docs.frigate.video/configuration/object_detectors/#openvino-detector

[tough yarrow]

as in this: detectors:
ov_0:
type: openvino
device: GPU
ov_1:
type: openvino
device: GPU

[mint wagon]

detectors:
  ov:
    type: openvino
    device: GPU

model:
  width: 300
  height: 300
  input_tensor: nhwc
  input_pixel_format: bgr
  path: /openvino-model/ssdlite_mobilenet_v2.xml
  labelmap_path: /openvino-model/coco_91cl_bkgr.txt

[tough yarrow]

thanks

[tough yarrow]

ffs I have broken everything... spent so many hours getting this setup the first time

what is wrong now

all I wanted to do was see if I could lower my inference time

now I've had to start from scratch and cant get anything working

not sure what it is about Frigate that is such a nightmare, I have no problem with any other dockers

managed to get it back on

no GPU stats anymore but low inference speed