[HW Accel Support]: Coral is very slow

[SebOM]

Describe the problem you are having

My Coral is very slow (130ms)

Version

0.14.1-f4f3cfa

Frigate config file

mqtt:
  enabled: true
  user: jeedom
  password: XKCuYAiiQNLnHavVPAVGO5b9yqljIMa3C29KN4ffddZNVQa79UdO0g9Mdbiq09tA
  host: 192.168.1.2
  port: 1883

ffmpeg:
  hwaccel_args: 
    preset-vaapi
  output_args:
    record: preset-record-generic-audio-aac

detectors:
   coral:
    type: edgetpu
    device: usb
    detect:
      enabled: true

go2rtc:
  streams:
    Salon:
      - rtsp://freeboxcam:[email protected]/live
      - ffmpeg:rtsp_entree#audio=opus#audio=aac
    JardinHD:
      - rtsp://motion:[email protected]:554/live/stream1
      - ffmpeg:rtsp_entree#audio=opus#audio=aac

timestamp_style:
  #Yes, the legacy software is capable of showing the timestamp, but i like all the configurations to be the same place
  position: tl
  format: '%d/%m/%Y %H:%M:%S'
  color:
    red: 255
    green: 255
    blue: 255
  thickness: 3

snapshots:
  enabled: true

cameras:
  Salon:
    ffmpeg:
      inputs:
        - path: rtsp://freeboxcam:[email protected]/live
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/Salon
          input_args: preset-rtsp-restream
          roles:
            - detect
            
    record:
      enabled: true
      retain:
        days: 14
        mode: all
      events:
        retain:
          default: 14
          mode: motion
        pre_capture: 3
        post_capture: 3
    detect:
      enabled: true
      width: 1280
      height: 720
      fps: 5

    objects:
      track:
        - person
      filters:
        person:
          min_score: 0.7
          threshold: 0.75
    motion:
      threshold: 30
      contour_area: 10
      improve_contrast: 'true'

  Jardin:
    ffmpeg:
      inputs:
        - path: rtsp://motion:[email protected]:554/live/stream1
          roles:
            - record
          #input_args: preset-rtsp-generic
        - path: rtsp://127.0.0.1:8554/JardinHD
          input_args: preset-rtsp-restream
          roles:
            
            - detect
    record:
      enabled: true
      retain:
        days: 14
        mode: all
      events:
        retain:
          default: 14
          mode: motion
        pre_capture: 3
        post_capture: 3
    detect:
      enabled: true
      fps: 5
    objects:
      track:
        - person
      mask: 0.214,0.178,0.213,0.098,0.188,0,0.215,0,0.248,0.107,0.248,0.197
      filters:
        person:
          min_score: 0.7
          threshold: 0.75
    motion:
      threshold: 30
      contour_area: 10
      improve_contrast: 'true'
      mask: 
        0.688,0.08,1,0.233,1,0,0.342,0,0.177,0.047,0.327,0.021,0.346,0,0.34,0,0,0,0,0.049,0,0.111,0.309,0,0.192,0,0.212,0.036,0.34,0.009
model:
  path: /cpu_model.tflite
version: 0.14

docker-compose file or Docker CLI command

services:

  frigate:
    container_name: frigate
    privileged: true
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:stable
    shm_size: "512mb" # update for your cameras based on calculation above
    devices:
      - /dev/bus/usb:/dev/bus/usb
      - /dev/dri/renderD128 # for intel hwaccel, needs to be updated for your hardware
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - /opt/frigate/config:/config
      - /mnt/frigate:/media/frigate
      - type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    ports:
      - "5000:5000"
      - "1935:1935" # RTMP feeds
      - "8971:8971"
      - "8554:8554" # RTSP feeds
      - "8555:8555/tcp" # WebRTC over tcp
      - "8555:8555/udp" # WebRTC over udp

Relevant Frigate log output

2025-01-20 08:52:06
frigate.detectors.plugins.edgetpu_tfl
TPU found

Relevant go2rtc log output

2025-01-20 08:51:59.242888071  [INFO] Preparing new go2rtc config...
2025-01-20 08:51:59.794575840  [INFO] Starting go2rtc...
2025-01-20 08:51:59.914874756  08:51:59.914 INF go2rtc platform=linux/amd64 revision=b2399f3 version=1.9.2
2025-01-20 08:51:59.914879832  08:51:59.914 INF config path=/dev/shm/go2rtc.yaml
2025-01-20 08:51:59.915392820  08:51:59.915 INF [rtsp] listen addr=:8554
2025-01-20 08:51:59.915488602  08:51:59.915 INF [api] listen addr=:1984
2025-01-20 08:51:59.915856282  08:51:59.915 INF [webrtc] listen addr=:8555/tcp
2025-01-20 08:52:09.139553028  [INFO] Starting go2rtc healthcheck service...

FFprobe output from your camera

No error

Install method

Proxmox via Docker

Object Detector

Coral

Network connection

Wired

Camera make and model

TAPO & FreeBOX

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response

[hawkeye217]

You should not be using the CPU model. Remove

model:
  path: /cpu_model.tflite

from your config.

[SebOM]

Hi,

Ok thx Detector Inference Speed are 10ms ;-)

[hawkeye217]

Great, glad it's working for you.

[arprv]

I've been struggling with this for over a month. I've bought and tried multiple USB3 PCIE cards in my Dell R720. Many, many cables, including the ones to power the PCIE card itself. I tried every PCIE slot there is. I tried powered hubs... Turns out I just copied the CPU detector model ID instead of the edgetpu one. There really should be an entry on "Troubleshooting EdgeTPU" page for this. An easy mistake to make when copying over a CPU detector config.

[arprv]

To elaborate, the model ID I was referring to is for Frigate+.

My inference speed went down from ~250ms to <8ms after switching to the edgetpu detector model ID.