[Detector Support]: Coral M-2 TPU under HassOS

[J-C-P]

Describe the problem you are having

I can’t get my Coral TPU to be recognized by Frigate
I still get No EdgeTPU was detected.

I believe I read all the post and all the discussion I can find about this kind of error. Obviously, I missed something, somewhere...

Version

0.14.1-f4f3cfa (Full Access)

Frigate config file

mqtt:
  host: 10.10.111.254 #Insert the IP address of your Home Assistant
  port: 1883 #Leave as default 1883 or change to match the port set in yout MQTT Broker configuration
  topic_prefix: frigate
  client_id: frigate
  user: mqtt_user #Change to match the username set in your MQTT Broker
  password: ********** #Change to match the password set in your MQTT Broker
  stats_interval: 60

detectors:
  coral:
    type: edgetpu
    device: pci

ffmpeg:
  hwaccel_args: preset-rpi-64-h264

cameras:
  Entree: #Change this to a name of the camera that makes sense to you
    ffmpeg:
      output_args:
        record: preset-record-generic-audio-aac #Insert this if your camera supports audio output
      inputs:
        - path: rtsp://administrator:*************@10.10.111.101:554/stream1 #Change as per your camera documentation for high quality feed
          roles:
            - record
        - path: rtsp://administrator:*************@10.10.111.101:554/stream2 #Change as per your camera documentation for low quality feed
          roles:
            - detect
    onvif:
      host: 10.10.111.101
      port: 2020
      user: administrator
      password: *************
    detect:
      height: 360 #Change this to match the resolution of your detection channel (in this case channel 1)
      width: 640 #Change this to match the resolution of your detection channel (in this case channel 1)
      fps: 10 #This is the frame rate for detection, between 5-10 fps is sufficient.

    record:
      enabled: true
      retain:
        days: 3 #The number of days a recording will be kept for after a motion is detected.
        mode: motion
      events:
        retain:
          default: 30 #The event data will be kept for 1 day. After this period, the event data will be automatically deleted.
          mode: motion
    objects:
      track:
        - person
        - car

docker-compose file or Docker CLI command

docker run \
  --name "/addon_ccab4aaf_frigate-fa" \
  --runtime "runc" \
  --mount type=bind,source=/dev,destination=/dev,readonly \
  --mount type=bind,source=/mnt/data/supervisor/addons/data/ccab4aaf_frigate-fa,destination=/data \
  --mount type=bind,source=/mnt/data/supervisor/homeassistant,destination=/config \
  --mount type=bind,source=/mnt/data/supervisor/media,destination=/media,bind-propagation=rslave \
  --log-driver "journald" \
  --log-opt tag="{{.Name}}" \
  --restart "no" \
  --add-host "hassio:172.30.32.2" \
  --add-host "supervisor:172.30.32.2" \
  --cap-add "PERFMON" \
  --publish "0.0.0.0:8555:8555/tcp" \
  --publish "0.0.0.0:8555:8555/udp" \
  --network "hassio" \
  --network-alias "ccab4aaf-frigate-fa" \
  --hostname "ccab4aaf-frigate-fa" \
  --domainname "local.hass.io" \
  --expose "5000/tcp" \
  --expose "8554/tcp" \
  --expose "8555/tcp" \
  --expose "8555/udp" \
  --env "CONFIG_FILE=/config/frigate.yml" \
  --env "TZ=Europe/Brussels" \
  --env "SUPERVISOR_TOKEN=5f00844df6765405eac9c654751d7c8e1540702f6bcaa6698fa2485aceb3842afcca8a6b7f1f85baa5e42326654fc58a807ef6688476a590" \
  --env "HASSIO_TOKEN=5f00844df6765405eac9c654751d7c8e1540702f6bcaa6698fa2485aceb3842afcca8a6b7f1f85baa5e42326654fc58a807ef6688476a590" \
  --env "PATH=/usr/lib/btbn-ffmpeg/bin:/usr/local/go2rtc/bin:/usr/local/tempio/bin:/usr/local/nginx/sbin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin" \
  --env "NVIDIA_VISIBLE_DEVICES=all" \
  --env "NVIDIA_DRIVER_CAPABILITIES=compute,video,utility" \
  --env "S6_LOGGING_SCRIPT=T 1 n0 s10000000 T" \
  --env "S6_CMD_WAIT_FOR_SERVICES_MAXTIME=0" \
  --label "supervisor_managed"="" \
  --detach \
  --entrypoint "/init" \
  "ghcr.io/blakeblackshear/frigate:0.14.1" \

Relevant Frigate log output

2024-11-08 10:42:37.581556851  [2024-11-08 11:42:37] detector.coral                 INFO    : Starting detection process: 206
2024-11-08 10:42:37.581643333  Process detector:coral:
2024-11-08 10:42:37.582625925  Traceback (most recent call last):
2024-11-08 10:42:37.583068314  [2024-11-08 11:42:37] frigate.detectors.plugins.edgetpu_tfl INFO    : Attempting to load TPU as pci
2024-11-08 10:42:37.583117389    File "/usr/lib/python3/dist-packages/tflite_runtime/interpreter.py", line 160, in load_delegate
2024-11-08 10:42:37.583119018      delegate = Delegate(library, options)
2024-11-08 10:42:37.583137166    File "/usr/lib/python3/dist-packages/tflite_runtime/interpreter.py", line 119, in __init__
2024-11-08 10:42:37.583138277      raise ValueError(capture.message)
2024-11-08 10:42:37.583162333  ValueError
2024-11-08 10:42:37.583183518  
2024-11-08 10:42:37.583184907  During handling of the above exception, another exception occurred:
2024-11-08 10:42:37.583185500  
2024-11-08 10:42:37.583198222  Traceback (most recent call last):
2024-11-08 10:42:37.583229074    File "/usr/lib/python3.9/multiprocessing/process.py", line 315, in _bootstrap
2024-11-08 10:42:37.583229981      self.run()
2024-11-08 10:42:37.583242537    File "/usr/lib/python3.9/multiprocessing/process.py", line 108, in run
2024-11-08 10:42:37.583243574      self._target(*self._args, **self._kwargs)
2024-11-08 10:42:37.583255722    File "/opt/frigate/frigate/object_detection.py", line 102, in run_detector
2024-11-08 10:42:37.583256926      object_detector = LocalObjectDetector(detector_config=detector_config)
2024-11-08 10:42:37.583269444    File "/opt/frigate/frigate/object_detection.py", line 53, in __init__
2024-11-08 10:42:37.583270518      self.detect_api = create_detector(detector_config)
2024-11-08 10:42:37.583282629    File "/opt/frigate/frigate/detectors/__init__.py", line 18, in create_detector
2024-11-08 10:42:37.583283555      return api(detector_config)
2024-11-08 10:42:37.583296074    File "/opt/frigate/frigate/detectors/plugins/edgetpu_tfl.py", line 41, in __init__
2024-11-08 10:42:37.583297277      edge_tpu_delegate = load_delegate("libedgetpu.so.1.0", device_config)
2024-11-08 10:42:37.583309111    File "/usr/lib/python3/dist-packages/tflite_runtime/interpreter.py", line 162, in load_delegate
2024-11-08 10:42:37.583310315      raise ValueError('Failed to load delegate from {}\n{}'.format(
2024-11-08 10:42:37.583326185  ValueError: Failed to load delegate from libedgetpu.so.1.0
2024-11-08 10:42:37.583326963  
2024-11-08 10:42:37.587672334  [2024-11-08 11:42:37] frigate.detectors.plugins.edgetpu_tfl ERROR   : No EdgeTPU was detected. If you do not have a Coral device yet, you must configure CPU detectors.

Install method

HassOS Addon

Object Detector

Coral

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

In order to access the UI's metrics pages, I needed to revert the detectors config to remove Coral config.

[NickM-27]

Do you see the coral in the hardware list in HA UI?

[J-C-P]

Hi,
Thanks for your answer.
Yes, I do.

[NickM-27]

what addon are you running specifically (what variant)

[J-C-P]

The Frigate (Full Access) Current version: 0.14.1

[NickM-27]

if you get a shell in the container and ls /dev/ what is shown?

[J-C-P]

[NickM-27]

that's very odd, I am not sure why it would fail to load, unless something else was actively using the coral

[J-C-P]

I'm kind of relieved it is not obvious... I'm on that since a few days...
Is there a way to detect if something else uses the Coral ?

[J-C-P]

Directly in a Raspberry Pi5.
I don't see any addon that could use the Coral.

[J-C-P]

I just tried to stop all other addons and try again with the detectors config. Same error. I reverted to "no detector" config :-/

[NickM-27]

yes, I don't know why this would happen unless there is some power issue causing the coral to be unable to fully start up

[J-C-P]

I will try with another power supply. It will have to wait a few hours for that.
Thanks for your support !
Have a good weekend :-)

[J-C-P]

I tried with another power supply. It was not the cause of the problem.
The problem is still there :-/

[J-C-P]

Is there something else I can try to find the issue ?

[NickM-27]

I'm not sure why it wouldn't work in that case. Can always try the normal addon instead of full access

[J-C-P]

Hi,
I tested the normal addon and I had the same result.
I also tested the beta version with the same result.
I confirm the "apex_0" device is present in both cases.

Is there a way to validate that the Coral works as expected ?
Preferably in the HassOS to avoid any docker issue.
Thanks

[NickM-27]

There are coral test repos, not sure if there is an addon for that

[J-C-P]

OK, thanks.
I will try to find something over there.

It is going to make me crazy !
I feel like it is something simple I missed but I can't find the detail I failed :-/

[J-C-P]

Youhouuu :-)

2024-11-12 18:30:42.399790832  [2024-11-12 19:30:42] frigate.detectors.plugins.edgetpu_tfl INFO    : Attempting to load TPU as pci
2024-11-12 18:30:42.406247312  [2024-11-12 19:30:42] frigate.detectors.plugins.edgetpu_tfl INFO    : TPU found

The issue was due to a Kernel update in Raspberry Pi5.
The solution is here under.
https://forums.raspberrypi.com/viewtopic.php?t=363682&sid=cb59b026a412f0dc041595951273a9ca&start=25

Add

dtoverlay=pciex1-compat-pi5,no-mip
dtoverlay=pcie-32bit-dma-pi5

to config.txt file.

I also added

dtoverlay=pciex1_gen=3

to force gen3 speed on my NVMe SSD. Not sure it changes something here.

[NickM-27]

Thanks for coming back with the solution