[Support]: i7-8550u proof of concept

[maxdd]

Hello,
I would like to setup a poc to see how frigate performs in my environment.
I have an i7-8550u with debian and docker as my main device and 1-2 Tapo C320.
Here, some hw info

ls /dev/dri/
by-path  card0  card1  renderD128  renderD129

drm_info -j | jq 'with_entries(.value |= .driver.desc)'
{
  "/dev/dri/card1": "Intel Graphics",
  "/dev/dri/card0": "nVidia Riva/TNT/GeForce/Quadro/Tesla/Tegra K1+"
}

*-display                 
       description: VGA compatible controller
       product: UHD Graphics 620 [8086:5917]
       vendor: Intel Corporation [8086]
       physical id: 2
       bus info: pci@0000:00:02.0
       logical name: /dev/fb0
       version: 07
       width: 64 bits
       clock: 33MHz
       capabilities: pciexpress msi pm vga_controller bus_master cap_list rom fb
       configuration: depth=32 driver=i915 latency=0 mode=1920x1080 resolution=1920,1080 visual=truecolor xres=1920 yres=1080
       resources: iomemory:1f0-1ef iomemory:1f0-1ef irq:142 memory:1ff0000000-1ff0ffffff memory:1fe0000000-1fefffffff ioport:5000(size=64) memory:c0000-dffff
  *-display
       description: 3D controller
       product: GM108M [GeForce 930MX] [10DE:134E]
       vendor: NVIDIA Corporation [10DE]
       physical id: 0
       bus info: pci@0000:01:00.0
       version: a2
       width: 64 bits
       clock: 33MHz
       capabilities: pm msi pciexpress bus_master cap_list rom
       configuration: driver=nouveau latency=0
       resources: irq:141 memory:b2000000-b2ffffff memory:a0000000-afffffff memory:b0000000-b1ffffff ioport:4000(size=128)

I've seen i have to point to the proper renderDxyz in the docker and i was wondering whether i should use D128 or D129 or cardX.
With openVino i have 15ms inference speed but i'm wondering whether going for the GPU would be better

[NickM-27]

in the vast majority of cases it is D128.

[NickM-27]

although i only see jsmpeg in the live (with terrible quality). I've tried this as well

this is because the naming is incorrect. You called the camera Tapo_C320_HD but you called the stream tapo_C320_hd

[maxdd]

The camera is called Tapo_C320 in the new config, i thought the path was what it matters.
What it's not clear though is if i have two streams

tapo_c320_hd 
tapo_c320_sd

how can i have 1 camera with two inputs if the camera must have the same name of the stream?

[NickM-27]

the stream that is used for the live view is the one that matches the name of the camera. so really, you would want it to look like this

go2rtc:
  streams:
    tapo_c320:
      - rtsp://itsmaxdd:C@[email protected]/stream1
    tapo_c320_sd:
      - rtsp://itsmaxdd:C@[email protected]/stream2

cameras:
  tapo_c320:
    enabled: True
    ffmpeg:
      hwaccel_args: preset-vaapi
      inputs:
        - path: rtsp://192.168.1.52:8554/tapo_c320_sd
          roles:
            - detect
        - path: rtsp://192.168.1.52:8554/tapo_c320
          roles:
            - record

[maxdd]

That indeed solved the issue, now i see both webrtc and mse.
Btw at startup i also receive

frigate/frigate/config.py

Line 1306 in 6476f8a

"Customizing more than a detector model path is unsupported."

which feels related to

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

is that normal? feels like labelmap is {}

[NickM-27]

Yes it's normal, it's but affecting your config

[aburow]

Using either GPU will improve performance of ffmpeg with hardware encode and decode.

[Waank1]

I also have an i7-8550u running VirtualBox under windows 10, have not been able to get any hwaccel to work, have read several posts and tried:
ffmpeg:
hwaccel_args: preset-vaapi

among others, figure it is because of the VM...

[maxdd]

Yes maybe with gpu-passthrough it can work?
Although Linux + docker is by far the best solution here