[Detector Support]: Crashing when using two coral edge tpu

[Parley33]

Describe the problem you are having

Hi, I changed the gpu on my server from rx 570 to gtx 1070, I created a new proxmox lxc, in lxc I have docker, and in docker I have frigate. It works really well, but today, when I did it and installed the frigate -0.15.1-tensorrt image, the card is visible and decodes correctly, but the dual edge tpu is detected only one by one. I use the adapter recommended for dual coral edge tpu and both on the host and in lxc itself are visible /dev/apex_0
/dev/apex_1
but after adding the detectors to the config I get this error in the logs and frigate restarts constantly:

Version

0.15.0-6cb5cfb

Frigate config file

mqtt:
  # Optional: Enable mqtt server (default: shown below)
  enabled: true
  # Required: host name
  host: 192.168.0.18
  # Optional: port (default: shown below)
  port: 1883
  # Optional: topic prefix (default: shown below)
  # NOTE: must be unique if you are running multiple instances
  topic_prefix: frigate
  # Optional: client id (default: shown below)
  # NOTE: must be unique if you are running multiple instances
  client_id: frigate
  # Optional: user
  # NOTE: MQTT user can be specified with an environment variable or docker secrets that must begin with 'FRIGATE_'.
  #       e.g. user: '{FRIGATE_MQTT_USER}'
  user: mqtt-user
  # Optional: password
  # NOTE: MQTT password can be specified with an environment variable or docker secrets that must begin with 'FRIGATE_'.
  #       e.g. password: '{FRIGATE_MQTT_PASSWORD}'
  password: pass



ffmpeg:
  output_args:
    record: preset-record-generic-audio-aac
  input_args: preset-rtsp-restream
  hwaccel_args: preset-nvidia


audio:
  enabled: true
  listen:
    - bark
    - fire_alarm
    - scream
    - speech
    - yell

go2rtc:
  streams:
    Rog_PTZ:
      - rtsp://admin:[email protected]/Preview_01_main
      - ffmpeg:back#audio=aac#video=h264#hardware



    Rog_PTZ_sub:
      - rtsp://admin:[email protected]/Preview_01_sub
      - ffmpeg:back#audio=aac

    Ogrod:
      - rtsp://admin:[email protected]/Preview_01_main
      - ffmpeg:back#audio=aac

    Ogrod_sub:
      - rtsp://admin:[email protected]/Preview_01_sub
      - ffmpeg:back#audio=aac

  webrtc:
    candidates:
      - 192.168.0.106:8555
      - stun:8555

detectors:
  coral1:
    type: edgetpu
    device: pci:0
  # coral2:
  #   type: edgetpu
  #   device: pci:1



record:
  enabled: true
  retain:
    days: 14
    mode: all
  preview:
    # Optional: Quality of recording preview (default: shown below).
    # Options are: very_low, low, medium, high, very_high
    quality: high

  alerts:
    retain:
      days: 12
  detections:
    retain:
      days: 12
objects:
  # Optional: list of objects to track from labelmap.txt (default: shown below)
  track:
    - person
    - car
    - cat
    - dog

  filters:
    person:
      # Optional: minimum width*height of the bounding box for the detected object (default: 0)
      min_area: 4000
      # Optional: maximum width*height of the bounding box for the detected object (default: 24000000)
      # max_area: 1050000
      # Optional: minimum decimal percentage for tracked object's computed score to be considered a true positive (default: shown below)
      threshold: 0.75
    car:
      # Optional: minimum width*height of the bounding box for the detected object (default: 0)
      min_area: 15000
      # Optional: maximum width*height of the bounding box for the detected object (default: 24000000)
      #max_area: 1000000
      # Optional: minimum decimal percentage for tracked object's computed score to be considered a true positive (default: shown below)
      threshold: 0.75
    dog:
      # Optional: minimum width*height of the bounding box for the detected object (default: 0)
      min_area: 4000
      # Optional: maximum width*height of the bounding box for the detected object (default: 24000000)
      #max_area: 1000000
      # Optional: minimum decimal percentage for tracked object's computed score to be considered a true positive (default: shown below)
      threshold: 0.75





cameras:
  Rog_PTZ: # <------ Name the camera
    enabled: true
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: rtsp://127.0.0.1:8554/Rog_PTZ

          input_args: preset-rtsp-restream
          roles:
            - record
            - detect


        # Low Resolution Stream
        - path: rtsp://127.0.0.1:8554/Rog_PTZ_sub
          input_args: preset-rtsp-restream
          roles:
            - audio


    detect:
      width: 2560
      height: 1440
      #width: 3840
      #height: 2160
      fps: 5
    snapshots:
      enabled: true

    onvif:
      host: 192.168.0.7
      port: 8000
      user: admin
      password: pass

    motion:
      mask:
        - 0.807,0,1,0,1,0.037,0.803,0.032
        - 0.002,0.211,0.086,0.196,0.316,0.106,0.349,0.053,0.339,0,0.002,0
    zones:
      MonitoringRog:
        coordinates: 0,0,0.003,0.995,0.998,0.994,1,0.036,0.799,0.041,0.798,0
        loitering_time: 0
        inertia: 3
        objects:
          - car
          - cat
          - dog
          - person
    review:
      alerts:
        required_zones: MonitoringRog
      detections:
        required_zones: MonitoringRog
    objects:
      filters:
        person:
          mask: 0.202,0.159,0.277,0.156,0.278,0.262,0.205,0.273
      mask:
        - 0.472,0.255,0.494,0.254,0.499,0.33,0.476,0.334
        - 0.887,0.325,0.921,0.323,0.923,0.445,0.889,0.438
  Ogrod: # <--- this will be changed to your actual camera later
    enabled: true
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: rtsp://127.0.0.1:8554/Ogrod
          input_args: preset-rtsp-restream
          roles:
            - record
            - detect

        # Low Resolution Stream
        - path: rtsp://127.0.0.1:8554/Ogrod_sub
          input_args: preset-rtsp-restream
          roles:
            - audio



    detect:
      width: 2560
      height: 1440
      fps: 5
    snapshots:
      enabled: true


    motion:
      mask:
        - 0.701,0.043,0.999,0.042,0.995,0,0.702,0
        - 0,0.158,0.134,0.153,0.256,0.174,0.28,0.16,0.314,0.12,0.358,0.083,0.419,0.082,0.451,0.083,0.418,0.129,0.419,0.223,0.421,0.254,0.352,0.259,0.262,0.275,0.217,0.238,0.19,0.194,0.179,0.183,0.148,0.217,0.085,0.31,0,0.283
        - 1,0.429,1,0.316,0.914,0.3,0.889,0.259,0.77,0.296,0.767,0.129,0.778,0.083,0.795,0,1,0
    zones:
      Drzwi:
        coordinates: 0.36,0.265,0.56,0.26,0.559,0.415,0.476,0.406,0.443,0.415,0.36,0.424
        loitering_time: 0
        inertia: 3
        objects: person
      Brama:
        coordinates: 0.939,0.31,1,0.319,0.995,0.509,0.768,0.412,0.768,0.295
        loitering_time: 0
        objects:
          - person
          - cat
          - dog
          - car
        inertia: 3
      Chodnik_drzwi_lewo:
        coordinates: 0.576,0.259,0.627,0.268,0.629,0.434,0.562,0.417,0.564,0.257
        loitering_time: 0
        inertia: 3
      Chodnik_Drzwi_lewo_poczatek:
        coordinates: 0.636,0.273,0.742,0.292,0.74,0.422,0.689,0.434,0.633,0.421
        loitering_time: 0
        inertia: 3
      Srodek_ogrodu:
        coordinates: 
          0,0.604,0,0.545,0,0.524,0,0.512,0,0.511,0,0.305,0.237,0.28,0.354,0.265,0.354,0.435,0.388,0.43,0.417,0.433,0.602,0.442,0.518,0.47,0.368,0.613,0.323,0.662,0.247,0.761,0.136,0.997,0,0.997
        loitering_time: 0
        objects:
          - car
          - person
          - dog
          - cat
        inertia: 3
      Przed_misiem:
        coordinates: 
          0.53,0.485,0.625,0.465,0.756,0.435,0.925,0.497,0.951,0.524,1,0.567,1,0.692,1,0.824,1,0.84,1,1,0.776,1,0.566,1,0.558,0.957,0.335,0.925,0.247,0.889,0.431,0.624,0.457,0.582
        loitering_time: 0
        inertia: 3
    review:
      alerts:
        required_zones:

          - Brama
          - Przed_misiem
          - Srodek_ogrodu
          - Drzwi
          - Chodnik_Drzwi_lewo_poczatek
          - Chodnik_drzwi_lewo
      detections:
        required_zones:
          - Brama
          - Przed_misiem
          - Drzwi
          - Chodnik_Drzwi_lewo_poczatek
          - Chodnik_drzwi_lewo
          - Srodek_ogrodu
    objects:
      mask:
        - 0.472,0.358,0.488,0.355,0.488,0.401,0.468,0.401
        - 0.208,0.418,0.124,0.441,0.126,0.407,0.207,0.391
        - 0.718,0.352,0.702,0.356,0.707,0.415,0.72,0.416
        - 0.375,0.288,0.409,0.287,0.41,0.343,0.378,0.336
        - 0.097,0.421,0.047,0.437,0.042,0.303,0.095,0.299
version: 0.15-1
camera_groups:
  Kamery:
    order: 1
    icon: LuActivity
    cameras:
      - Rog_PTZ
      - Ogrod
logger:
  default: info

semantic_search:
  enabled: false
  reindex: false
  model_size: small
notifications:
  enabled: 'false'
  email: [email protected]


telemetry:
    # Optional: Configure system stats
  stats:
    # Enable AMD GPU stats (default: shown below)
    amd_gpu_stats: true
    # Enable Intel GPU stats (default: shown below)
    intel_gpu_stats: true
    # Enable network bandwidth stats monitoring for camera ffmpeg processes, go2rtc, and object detectors. (default: shown below)
    # NOTE: The container must either be privileged or have cap_net_admin, cap_net_raw capabilities enabled.
    network_bandwidth: true

docker-compose file or Docker CLI command

services:
  frigate:
    container_name: frigate
    privileged: true # this may not be necessary for all setups
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:0.15.1-tensorrt
    deploy:
      # <------------- Add this section
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities:
                - gpu
    shm_size: 1024mb # update for your cameras based on calculation above
    devices:
      - /dev/bus/usb:/dev/bus/usb 
      - /dev/apex_0:/dev/apex_0 
      - /dev/apex_1:/dev/apex_1
      - /dev/video11:/dev/video11 
      - /dev/dri/renderD128:/dev/dri/renderD128 
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - ./config:/config
      - /MonitoringNagrania:/media/frigate
      - type: tmpfs
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    ports:
      - 8971:8971
      - 5100:5000 # Internal unauthenticated access. Expose carefully.
      - 8554:8554 # RTSP feeds
      - 8555:8555/tcp # WebRTC over tcp
      - 8555:8555/udp # WebRTC over udp
    environment:
      FRIGATE_RTSP_PASSWORD: password
      PLUS_API_KEY: key
networks:
  dockge_default:
    external: true

Relevant Frigate log output

2025-06-06 16:48:42.869404818  [INFO] Preparing Frigate...
2025-06-06 16:48:43.011340564  [INFO] Starting Frigate...
2025-06-06 16:48:45.901712416  [2025-06-06 16:48:45] frigate.util.config            INFO    : Checking if frigate config needs migration...
2025-06-06 16:48:45.963684998  [2025-06-06 16:48:45] frigate.util.config            INFO    : frigate config does not need migration...
2025-06-06 16:48:46.127689239  [2025-06-06 16:48:46] frigate.app                    INFO    : Starting Frigate (0.15.0-6cb5cfb)
2025-06-06 16:48:46.138190589  [2025-06-06 16:48:46] peewee_migrate.logs            INFO    : Starting migrations
2025-06-06 16:48:46.138605079  [2025-06-06 16:48:46] peewee_migrate.logs            INFO    : There is nothing to migrate
2025-06-06 16:48:46.472579724  [2025-06-06 16:48:46] frigate.app                    INFO    : Recording process started: 357
2025-06-06 16:48:46.473759194  [2025-06-06 16:48:46] frigate.app                    INFO    : Review process started: 366
2025-06-06 16:48:46.477092025  [2025-06-06 16:48:46] frigate.app                    INFO    : go2rtc process pid: 103
2025-06-06 16:48:46.502266265  [2025-06-06 16:48:46] detector.coral1                INFO    : Starting detection process: 373
2025-06-06 16:48:46.506041313  [2025-06-06 16:48:46] detector.coral2                INFO    : Starting detection process: 376
2025-06-06 16:48:46.521216128  [2025-06-06 16:48:46] frigate.detectors.plugins.edgetpu_tfl INFO    : Attempting to load TPU as pci:0
2025-06-06 16:48:46.521328605  [2025-06-06 16:48:46] frigate.detectors.plugins.edgetpu_tfl INFO    : TPU found
2025-06-06 16:48:46.522898691  [2025-06-06 16:48:46] frigate.detectors.plugins.edgetpu_tfl INFO    : Attempting to load TPU as pci:1
2025-06-06 16:48:46.523001712  [2025-06-06 16:48:46] frigate.detectors.plugins.edgetpu_tfl ERROR   : No EdgeTPU was detected. If you do not have a Coral device yet, you must configure CPU detectors.
2025-06-06 16:48:46.523714247  [2025-06-06 16:48:46] frigate.app                    INFO    : Output process started: 410
2025-06-06 16:48:46.525783346  Process detector:coral2:
2025-06-06 16:48:46.525786494  Traceback (most recent call last):
2025-06-06 16:48:46.525788419    File "/usr/lib/python3/dist-packages/tflite_runtime/interpreter.py", line 160, in load_delegate
2025-06-06 16:48:46.525792547      delegate = Delegate(library, options)
2025-06-06 16:48:46.525795097    File "/usr/lib/python3/dist-packages/tflite_runtime/interpreter.py", line 119, in __init__
2025-06-06 16:48:46.525813618      raise ValueError(capture.message)
2025-06-06 16:48:46.525815246  ValueError
2025-06-06 16:48:51.265132  2025-06-06 16:48:46.525816484
2025-06-06 16:48:46.525818710  During handling of the above exception, another exception occurred:
2025-06-06 16:48:51.265176  2025-06-06 16:48:46.525820051
2025-06-06 16:48:46.525821434  Traceback (most recent call last):
2025-06-06 16:48:46.525823158    File "/usr/lib/python3.9/multiprocessing/process.py", line 315, in _bootstrap
2025-06-06 16:48:46.525842075      self.run()
2025-06-06 16:48:46.525843850    File "/opt/frigate/frigate/util/process.py", line 41, in run_wrapper
2025-06-06 16:48:46.525845411      return run(*args, **kwargs)
2025-06-06 16:48:46.525847003    File "/usr/lib/python3.9/multiprocessing/process.py", line 108, in run
2025-06-06 16:48:46.525877640      self._target(*self._args, **self._kwargs)
2025-06-06 16:48:46.525879278    File "/opt/frigate/frigate/object_detection.py", line 121, in run_detector
2025-06-06 16:48:46.525880955      object_detector = LocalObjectDetector(detector_config=detector_config)
2025-06-06 16:48:46.525882451    File "/opt/frigate/frigate/object_detection.py", line 68, in __init__
2025-06-06 16:48:46.525883947      self.detect_api = create_detector(detector_config)
2025-06-06 16:48:46.525885516    File "/opt/frigate/frigate/detectors/__init__.py", line 18, in create_detector
2025-06-06 16:48:46.525886835      return api(detector_config)
2025-06-06 16:48:46.525900183    File "/opt/frigate/frigate/detectors/plugins/edgetpu_tfl.py", line 41, in __init__
2025-06-06 16:48:46.525901861      edge_tpu_delegate = load_delegate("libedgetpu.so.1.0", device_config)
2025-06-06 16:48:46.525903490    File "/usr/lib/python3/dist-packages/tflite_runtime/interpreter.py", line 162, in load_delegate
2025-06-06 16:48:46.525905095      raise ValueError('Failed to load delegate from {}\n{}'.format(
2025-06-06 16:48:46.525906521  ValueError: Failed to load delegate from libedgetpu.so.1.0
2025-06-06 16:48:51.265189  2025-06-06 16:48:46.525907615
2025-06-06 16:48:46.567263877  [2025-06-06 16:48:46] frigate.app                    INFO    : Camera processor started for Rog_PTZ: 425
2025-06-06 16:48:46.582222242  [2025-06-06 16:48:46] frigate.app                    INFO    : Camera processor started for Ogrod: 426
2025-06-06 16:48:46.582295090  [2025-06-06 16:48:46] frigate.app                    INFO    : Capture process started for Rog_PTZ: 431
2025-06-06 16:48:46.630204171  [2025-06-06 16:48:46] frigate.app                    INFO    : Capture process started for Ogrod: 446
2025-06-06 16:48:46.658755143  [2025-06-06 16:48:46] frigate.audio_manager          INFO    : Audio processor started (pid: 449)
2025-06-06 16:48:46.852348780  [2025-06-06 16:48:46] frigate.api.fastapi_app        INFO    : Starting FastAPI app
2025-06-06 16:48:47.016009244  [2025-06-06 16:48:47] frigate.api.fastapi_app        INFO    : FastAPI started

Install method

Docker Compose

Object Detector

Coral

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

I used this tutorial for passthrough in lxc:
https://www.youtube.com/watch?v=lNGNRIJ708k

[NickM-27]

It would be good to see ls /dev/ run within the Frigate container

[Parley33]

[NickM-27]

I see, in that case I have no idea why they would not be recognized

[Parley33]

before the change I was using docker + portainer and now I'm using docker + dockge, maybe this change has some influence?

[NickM-27]

I would not expect it to, dockge is quite common and recommended. There is no other service, or perhaps old instance of frigate that could be stealing the coral connection, right?

[Parley33]

No, they are all turned off and disconnected from coral