[Detector Support]: New yolov8n config fails to start

[f1d094]

Describe the problem you are having

Tried to run with yolov8n models as described in object_detectors.md but frigate fails to start: "ValueError: Cannot set tensor: Got value of type UINT8 but expected type INT8 for input 0, name: serving_default_images:0"

Version

0.13.2-6476f8a

Frigate config file

model:
  labelmap_path: /config/model_cache/yolov8/labels.txt
  model_type: yolov8
  path: /config/model_cache/yolov8/yolov8n_320x320_edgetpu.tflite

detectors:
  coral1:
    type: edgetpu
    device: usb:0

ffmpeg:
  hwaccel_args: preset-nvidia-h264
  input_args: -fflags +genpts+igndts+discardcorrupt
  output_args:
    record: preset-record-generic-audio-copy

birdseye:
  enabled: true
  restream: true
  mode: motion
  height: 1920
  width: 1080
  layout:
    scaling_factor: 1

motion:
  #threshold: 22
  threshold: 50

objects:
  track:
  - person
  - bicycle
  - car 
  - dog 
  - cat 

record:
  enabled: true
  retain:
    days: 7
    mode: motion
  events:
    pre_capture: 2
    post_capture: 2
    retain:
      default: 60
      mode: active_objects

snapshots:
  enabled: true
  retain:
    default: 14

go2rtc:
  log:
    level: info
    api: info
    exec: info
    ngrok: info
    rtsp: info
    streams: info
    webrtc: info
  webrtc:
    candidates:
    - 127.0.0.1:8555

streams:

docker-compose file or Docker CLI command

podman run -d \
  --name frigate \
  --privileged \
  --memory=8g \
  --memory-swap=8g \
  --memory-swappiness=0 \
  --ipc=private \
  --restart unless-stopped \
  --mount type=tmpfs,target=/tmp/cache,tmpfs-size=1g \
  --device /dev/bus/usb:/dev/bus/usb \
  --device nvidia.com/gpu=0 \
  --security-opt=label=disable \
  --cap-add=CAP_PERFMON \
  --shm-size 256m \
  -v /mnt/Media:/media/frigate \
  -v /opt/Frigate/config:/config \
  -v /etc/localtime:/etc/localtime:ro \
  --ip=172.30.2.2 \
  -p 8555:8555/tcp \
  -p 8555:8555/udp \
  --network=daleks-net \
  -e DOWNLOAD_YOLOV8=1 \
  ghcr.io/blakeblackshear/frigate:stable

Relevant log output

2024-02-26T18:09:07.268998-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.268897497  Process detector:coral1:
2024-02-26T18:09:07.271086-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.270976278  Traceback (most recent call last):
2024-02-26T18:09:07.271226-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.270980649    File "/usr/lib/python3.9/multiprocessing/process.py", line 315, in _bootstrap
2024-02-26T18:09:07.271314-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.270982370      self.run()
2024-02-26T18:09:07.272451-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.271056406    File "/usr/lib/python3.9/multiprocessing/process.py", line 108, in run 
2024-02-26T18:09:07.272511-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.271058349      self._target(*self._args, **self._kwargs)
2024-02-26T18:09:07.272555-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.271059895    File "/opt/frigate/frigate/object_detection.py", line 125, in run_detector
2024-02-26T18:09:07.273063-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.271061416      detections = object_detector.detect_raw(input_frame)
2024-02-26T18:09:07.273121-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.271062913    File "/opt/frigate/frigate/object_detection.py", line 75, in detect_raw
2024-02-26T18:09:07.273161-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.271064407      return self.detect_api.detect_raw(tensor_input=tensor_input)
2024-02-26T18:09:07.273197-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.271066002    File "/opt/frigate/frigate/detectors/plugins/edgetpu_tfl.py", line 59, in detect_raw
2024-02-26T18:09:07.273233-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.271067631      self.interpreter.set_tensor(self.tensor_input_details[0]["index"], tensor_input)
2024-02-26T18:09:07.273267-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.271069249    File "/usr/lib/python3/dist-packages/tflite_runtime/interpreter.py", line 572, in set_tensor
2024-02-26T18:09:07.273302-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.271070676      self._interpreter.SetTensor(tensor_index, value)
2024-02-26T18:09:07.273341-08:00 testme frigate[1360348]: 2024-02-26 18:09:07.271072473  ValueError: Cannot set tensor: Got value of type UINT8 but expected type INT8 for input 0, name: serving_default_images:0

Operating system

Debian

Install method

Docker CLI

Coral version

USB

Any other information that may be helpful

At first I ran with two coral usb units plugged in...then I saw the "multiprocessing" error and thought it didn't like having two. Unplugged one and modified the config...but failed the same.

[NickM-27]

This is not supported in 0.13.2

[f1d094]

@NickM-27 Well that explains it. Maybe in 0.13.3?

[NickM-27]

No, it has been committed to dev so it will be in 0.14

[f1d094]

Will look forward to trying it in 0.14 then. Thank you for all your efforts. Tremendous project.

[Retic1337]

I am still having this error on 0.14.1 with a PCI edgetpu. TPU is working properly with stock configuration but has errors with yolov8n. i have downloaded multiple files to try as well ash installed ultralytics and created my own with no success.

[suprimex]

Is there any solution? Same problem with 0.17 as I see: ValueError: Cannot set tensor: Got value of type UINT8 but expected type INT8 for input 0, name: serving_default_images:0

[hawkeye217]

Frigate does not support YOLOv8 due to licensing issues: #10717