[HW Accel Support]: I cannot disable detection. I have VAAPI enabled and OPENVINO set to GPU, and also have Detect: disabled but my CPU is still getting hammered in OV and inference at 10ms.

[bobloadmire]

Describe the problem you are having

I cannot disable detection. I have VAAPI enabled and OPENVINO set to GPU, and also have Detect: disabled but my CPU is still getting hammered in OV and inference at 10ms.

Inference is showing 10ms and 150% cpu usage, even though its set to OPENVINO and forced on GPU. I have detect disabled, but my CPU is still getting hammered. I have no idea how to disable it or whats hammering my cpu. I have VAAPI enabled and the stream is 1080p,

FFprobe: [{"return_code":0,"stderr":"","stdout":{"programs":[],"streams":[{"avg_frame_rate":"15/1","codec_long_name":"H.264/AVC/MPEG-4AVC/MPEG-4part10","height":1080,"width":1920}]}}]

config:

mqtt:
  enabled: false
ffmpeg:
  hwaccel_args: preset-vaapi
  retry_interval: 10
detect:
  enabled: false
cameras:
  test1:
    enabled: false
    ffmpeg:
      inputs:
        - path: rtsp://192.168.0.4:8554/lyras-room
          #input_args: -re -stream_loop -1 -fflags +genpts
          #roles:
            #- detect
            #- rtmp
    detect:
      height: 1080
      width: 1920
      fps: 5
  test2:
    enabled: False
    ffmpeg:
      inputs:
        - path: rtsp://192.168.0.4:8554/kitchen
          #input_args: -re -stream_loop -1 -fflags +genpts
          #roles:
            #- detect
            #- rtmp
    detect:
      height: 1080
      width: 1920
      fps: 5
  test3:
    ffmpeg:
      inputs:
        - path: rtsp://192.168.0.4:8554/lincs-room
          #input_args: -re -stream_loop -1 -fflags +genpts
          roles:
            - detect
            #- rtmp
    detect:
      height: 1080
      width: 1920
      fps: 5
detectors:
  ov:
    type: openvino
    device: GPU
    model:
      path: /openvino-model/FP16/ssdlite_mobilenet_v2.xml
model:
  width: 300
  height: 300
  input_tensor: nhwc
  input_pixel_format: bgr
  labelmap_path: /openvino-model/coco_91cl_bkgr.txt
audio:
  # Optional: Enable audio events (default: shown below)
  enabled: False

Version

13.2

Frigate config file

mqtt:
  enabled: false
ffmpeg:
  hwaccel_args: preset-vaapi
  retry_interval: 10
detect:
  enabled: false
cameras:
  test1:
    enabled: false
    ffmpeg:
      inputs:
        - path: rtsp://192.168.0.4:8554/lyras-room
          #input_args: -re -stream_loop -1 -fflags +genpts
          #roles:
            #- detect
            #- rtmp
    detect:
      height: 1080
      width: 1920
      fps: 5
  test2:
    enabled: False
    ffmpeg:
      inputs:
        - path: rtsp://192.168.0.4:8554/kitchen
          #input_args: -re -stream_loop -1 -fflags +genpts
          #roles:
            #- detect
            #- rtmp
    detect:
      height: 1080
      width: 1920
      fps: 5
  test3:
    ffmpeg:
      inputs:
        - path: rtsp://192.168.0.4:8554/lincs-room
          #input_args: -re -stream_loop -1 -fflags +genpts
          roles:
            - detect
            #- rtmp
    detect:
      height: 1080
      width: 1920
      fps: 5
detectors:
  ov:
    type: openvino
    device: GPU
    model:
      path: /openvino-model/FP16/ssdlite_mobilenet_v2.xml
model:
  width: 300
  height: 300
  input_tensor: nhwc
  input_pixel_format: bgr
  labelmap_path: /openvino-model/coco_91cl_bkgr.txt
audio:
  # Optional: Enable audio events (default: shown below)
  enabled: False

docker-compose file or Docker CLI command

n/a

Relevant log output

[INFO] Preparing Frigate...
[INFO] Starting Frigate...
[2024-04-24 18:29:17] frigate.app                    INFO    : Starting Frigate (0.13.2-)
[2024-04-24 18:29:17] frigate.config                 WARNING : Customizing more than a detector model path is unsupported.
[2024-04-24 18:29:17] peewee_migrate.logs            INFO    : Starting migrations
[2024-04-24 18:29:17] peewee_migrate.logs            INFO    : There is nothing to migrate
[2024-04-24 18:29:17] frigate.app                    INFO    : Recording process started: 2492
[2024-04-24 18:29:17] frigate.app                    INFO    : go2rtc process pid: 128
[2024-04-24 18:29:17] frigate.app                    INFO    : Output process started: 2502
[2024-04-24 18:29:17] detector.ov                    INFO    : Starting detection process: 2501
[2024-04-24 18:29:17] frigate.app                    INFO    : Camera processor not started for disabled camera test1
[2024-04-24 18:29:17] frigate.app                    INFO    : Camera processor not started for disabled camera test2
[2024-04-24 18:29:17] frigate.app                    INFO    : Camera processor started for test3: 2508
[2024-04-24 18:29:17] frigate.app                    INFO    : Capture process not started for disabled camera test1
[2024-04-24 18:29:17] frigate.app                    INFO    : Capture process not started for disabled camera test2
[2024-04-24 18:29:17] frigate.app                    INFO    : Capture process started for test3: 2509
[2024-04-24 18:30:48] frigate.detectors.plugins.openvino INFO    : Model Input Shape: [1,300,300,3]
[2024-04-24 18:30:48] frigate.detectors.plugins.openvino INFO    : Model Output-0 Shape: [1,1,100,7]
[2024-04-24 18:30:48] frigate.detectors.plugins.openvino INFO    : Model has 1 Output Tensors

FFprobe output from your camera

[{"return_code":0,"stderr":"","stdout":{"programs":[],"streams":[{"avg_frame_rate":"15/1","codec_long_name":"H.264/AVC/MPEG-4AVC/MPEG-4part10","height":1080,"width":1920}]}}]

Operating system

Proxmox

Install method

Docker CLI

Network connection

Wired

Camera make and model

Wyze cam through Wyze Bridge.

Any other information that may be helpful

No response

[bobloadmire]

https://bobloadmire.duckdns.org/s/Aq6YawEy4ENoCfK

here's a screenshot of system. clearly shows that detection is disabled on both streams, but OV is slappin my cpus around still. And I have OPENVINO forced to GPU as well. none of this make sense to me.

[NickM-27]

You're using an unofficial direct lxc installation method if I remember correctly which is most likely the proboem

[NickM-27]

It could be so many different things and none of the Frigate maintainers are knowledgeable about proxmox as far as I know

[bobloadmire]

100% understand. I was thinking frigate might tell me hey x and y work but z is not, and I could track that down in proxmox.

[NickM-27]

frigate instructs ffmpeg to use the hardware, you could try running ffmpeg directly in debug mode and see if it complains about something not working

[bobloadmire]

if im understanding correctly, all ffmpeg logs are sent at error level, so I should have frigate report errors right? https://docs.frigate.video/configuration/advanced/

[NickM-27]

No, you'd want to increase the log level so you get more details. Frigate also only prints ffmpeg logs when the stream crashes which is why I said it should be run outside of frigate so the logs can be viewed immediately