[Detector Support]: yolov9(s or t) throws "f32 vs u8" error on 0.16.1

[JasonSwindle]

Describe the problem you are having

Trying out Frigate+ YOLOv9, s or t, base model or tuned model crash Frigate with an error about f32 vs u8. What am I doing wrong?

Version

0.16.1-e664cb2

Frigate config file

mqtt:
  host: mqtt

database:
  path: /config/frigate.db

detectors:
  ov_0:
    type: openvino
    device: GPU

ffmpeg:
  hwaccel_args: preset-vaapi

proxy:
  header_map:
    user: x-forwarded-user

auth:
  reset_admin_password: true

model:
  path: plus://<REMOVED>

#  width: 320
#  height: 320
#   input_tensor: nhwc
#   input_pixel_format: bgr
#   path: /openvino-model/ssdlite_mobilenet_v2.xml
#   labelmap_path: /openvino-model/coco_91cl_bkgr.txt

objects:
  track:
    - person
    - car
    - dog
    - cat
    - face
    - license_plate
    - ups
    - fedex
    - package
    - deer

record:
  sync_recordings: true
  enabled: true
  retain:
    days: 7
    mode: active_objects
  alerts:
      # Optional: Number of seconds before the detection to include (default: shown below)
    pre_capture: 10
      # Optional: Number of seconds after the detection to include (default: shown below)
    post_capture: 5
    retain:
      days: 7
      mode: active_objects
  detections:
    # Optional: Number of seconds before the detection to include (default: shown below)
    pre_capture: 10
    # Optional: Number of seconds after the detection to include (default: shown below)
    post_capture: 5
    retain:
      days: 7
      mode: active_objects

go2rtc:

  webrtc:
    candidates:
      - 10.0.0.98:8555
      - stun:8555

  streams:
    driveway:
      - ffmpeg:rtsp://frigate:{FRIGATE_DRIVEWAY_PASSWORD}@driveway.intra.swindle.casa:554/h264/ch1/main/av_stream#video=copy#hardware
    driveway_sub:
      - ffmpeg:rtsp://frigate:{FRIGATE_DRIVEWAY_PASSWORD}@driveway.intra.swindle.casa:554/H264/ch1/sub/av_stream#video=copy#hardware

    retaining_wall:
      - rtsp://admin:{FRIGATE_RETAINING_WALL_PASSWORD}@front-door.intra.swindle.casa:554/cam/realmonitor?channel=1&subtype=0
    retaining_wall_sub:
      - rtsp://admin:{FRIGATE_RETAINING_WALL_PASSWORD}@front-door.intra.swindle.casa:554/cam/realmonitor?channel=1&subtype=1

  homekit:
    driveway:
      name: Drive Way                   # Name displayed in HomeKit
      pin: 12345678                     # PIN to pair with HomeKit
      device_id: driveway               # Unique identifier for HomeKit
      device_private: driveway          # custom key, default: generated from stream ID
      pairings:
        - client_id=7602D9AC-5319-46F3-938F-E86F338BC2F6&client_public=0a8ec0a3e592199703710a1021886946343f0f78b074e0ddeb91ab24d4169ea8&permissions=1
        - client_id=F340B187-69DE-417E-A180-BACE795395AA&client_public=047805104ffcde26dd8f1f24881cc456dc30668860f6fab3609d3999ba5623e4&permissions=0

    retaining_wall:
      name: Retaining Wall              # Name displayed in HomeKit
      pin: 12345678                     # PIN to pair with HomeKit
      device_id: retaining_wall         # Unique identifier for HomeKit
      device_private: retaining_wall    # custom key, default: generated from stream ID
      pairings:
        - client_id=7602D9AC-5319-46F3-938F-E86F338BC2F6&client_public=0a8ec0a3e592199703710a1021886946343f0f78b074e0ddeb91ab24d4169ea8&permissions=1
        - client_id=F340B187-69DE-417E-A180-BACE795395AA&client_public=047805104ffcde26dd8f1f24881cc456dc30668860f6fab3609d3999ba5623e4&permissions=0

cameras:
  driveway:
    ffmpeg:
      output_args:
        record: preset-record-generic-audio-copy
      inputs:
        - path: rtsp://127.0.0.1:8554/driveway
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/driveway_sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    detect:
      enabled: true
      width: 1920
      height: 536
      fps: 10
    motion:
      threshold: 50
      contour_area: 10
      improve_contrast: true
    objects:
      filters:
        car:
          mask:
            0.226,0.765,0.23,0.65,0.679,0.79,1,0.673,1,1,0.681,0.997,0.433,1,0.326,1,0.286,0.847
    zones:
      Driveway:
        coordinates: 0.252,0,0.004,0.013,0,1,0.674,1,0.323,0.902,0.226,0.767
        loitering_time: 0
        inertia: 3
      Road:
        coordinates: 0.253,0,0.225,0.758,0.322,0.907,0.684,1,1,1,1,0
        loitering_time: 0
        inertia: 3
    review:
      detections:
        required_zones:
          - Driveway
          - Road
      alerts: {}
    live:
      streams:
        driveway: driveway
  retaining_wall:
    ffmpeg:
      output_args:
        record: preset-record-generic-audio-copy
      inputs:
        - path: rtsp://127.0.0.1:8554/retaining_wall
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/retaining_wall_sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    detect:
      enabled: true
      width: 640
      height: 480
      fps: 5
    objects:
      filters:
        car:
          mask:
            0,0.472,0,0,0.452,0,1,0,0.866,0.434,0.641,0.611,0.514,0.64,0.46,0.652,0.431,0.652,0.332,0.623,0.313,0.661,0.114,0.701,0,0.727
        package:
          mask:
            0.349,0.595,0.157,0.999,0.389,0.998,0.994,0.995,0.926,0.703,0.698,0.542,0.654,0.562
    zones:
      Driveway:
        coordinates:
          0.389,0.617,0.269,1,0.784,1,1,1,1,0.905,1,0.672,1,0.355,1,0.016,0.862,0.426,0.548,0.591,0.496,0.619,0.444,0.628,0.424,0.608
        loitering_time: 0
        inertia: 3
    review:
      detections:
        required_zones: Driveway
      alerts:
        required_zones: Driveway
    motion:
      threshold: 45
      improve_contrast: true
    live:
      streams:
        retaining_wall: retaining_wall
snapshots:
  enabled: true

telemetry:
  # Optional: Enabled network interfaces for bandwidth stats monitoring (default: empty list, let nethogs search all)
  network_interfaces: []
  # Optional: Configure system stats
  stats:
    # Enable AMD GPU stats (default: shown below)
    amd_gpu_stats: false
    # 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
  # Optional: Enable the latest version outbound check (default: shown below)
  # NOTE: If you use the HomeAssistant integration, disabling this will prevent it from reporting new versions
  version_check: false

# genai:
#   enabled: True
#   provider: gemini
#   api_key: "{FRIGATE_GEMINI_API_KEY}"
#   model: gemini-2.5-flash-preview-04-17

semantic_search:
  enabled: true
  reindex: false
  model_size: large

detect:
  enabled: true
face_recognition:
  enabled: true
  model_size: large
lpr:
  enabled: true
classification:
  bird:
    enabled: true
version: 0.16-0

docker-compose file or Docker CLI command

frigate:
    image: ghcr.io/blakeblackshear/frigate:0.16.1
    container_name: frigate
    hostname: frigate
    restart: on-failure
    network_mode: host
    environment:
      FRIGATE_RTSP_PASSWORD: ${FRIGATE_RTSP_PASSWORD}
      PLUS_API_KEY: ${PLUS_API_KEY}
      FRIGATE_GEMINI_API_KEY: ${FRIGATE_GEMINI_API_KEY}
      FRIGATE_DRIVEWAY_PASSWORD: ${FRIGATE_DRIVEWAY_PASSWORD}
      FRIGATE_RETAINING_WALL_PASSWORD: ${FRIGATE_RETAINING_WALL_PASSWORD}
      TZ: ${TIME_ZONE}
    shm_size: 2gb
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - /mnt/user/appdata/frigate:/config
      - /mnt/user/cctv/frigate:/media/frigate
    tmpfs:
      - /tmp/cache
    devices:
      - /dev/bus/usb:/dev/bus/usb
      - /dev/dri/renderD128:/dev/dri/renderD128
    stop_grace_period: 5m
    cap_add:
      - CAP_NET_ADMIN
      - CAP_NET_RAW
      - CAP_PERFMON
    depends_on:
      - mqtt

Relevant Frigate log output

2025-09-12 20:34:46.877609427  [2025-09-12 20:34:46] detector.ov_0                  INFO    : Starting detection process: 839
2025-09-12 20:34:47.900360876  Process detector:ov_0:
2025-09-12 20:34:47.900364018  Traceback (most recent call last):
2025-09-12 20:34:47.900365654    File "/usr/lib/python3.11/multiprocessing/process.py", line 314, in _bootstrap
2025-09-12 20:34:47.900366584      self.run()
2025-09-12 20:34:47.900367828    File "/opt/frigate/frigate/util/process.py", line 41, in run_wrapper
2025-09-12 20:34:47.900368863      return run(*args, **kwargs)
2025-09-12 20:34:47.900369983             ^^^^^^^^^^^^^^^^^^^^
2025-09-12 20:34:47.900371328    File "/usr/lib/python3.11/multiprocessing/process.py", line 108, in run
2025-09-12 20:34:47.900372466      self._target(*self._args, **self._kwargs)
2025-09-12 20:34:47.900373840    File "/opt/frigate/frigate/object_detection/base.py", line 136, in run_detector
2025-09-12 20:34:47.900374860      detections = object_detector.detect_raw(input_frame)
2025-09-12 20:34:47.900376037                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2025-09-12 20:34:47.900377043    File "/opt/frigate/frigate/object_detection/base.py", line 86, in detect_raw
2025-09-12 20:34:47.900378059      return self.detect_api.detect_raw(tensor_input=tensor_input)
2025-09-12 20:34:47.900379078             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2025-09-12 20:34:47.900380435    File "/opt/frigate/frigate/detectors/plugins/openvino.py", line 150, in detect_raw
2025-09-12 20:34:47.900381472      infer_request.infer(input_tensor)
2025-09-12 20:34:47.900382869    File "/usr/local/lib/python3.11/dist-packages/openvino/runtime/ie_api.py", line 132, in infer
2025-09-12 20:34:47.900383946      return OVDict(super().infer(_data_dispatch(
2025-09-12 20:34:47.900385051                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2025-09-12 20:34:47.900386369  RuntimeError: Exception from src/inference/src/cpp/infer_request.cpp:116:
2025-09-12 20:34:47.900387584  Exception from src/inference/src/cpp/infer_request.cpp:66:
2025-09-12 20:34:47.900389184  Check 'port.get_element_type() == tensor->get_element_type()' failed at src/plugins/intel_gpu/src/plugin/sync_infer_request.cpp:142:
2025-09-12 20:34:47.900390233  [GPU] Mismatch tensor and port type: f32 vs u8

Install method

Docker Compose

Object Detector

OpenVino

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response

[JasonSwindle]

I see in the docs:

warning
If you are using a Frigate+ YOLOv9 model, you should not define any of the below model parameters in your config except for path. See the Frigate+ model docs for more information on setting up your model.

If I define my config as:

model:
  path: plus://9ef45d9bed92e71b72ca560013078e80
  model_type: yolo-generic
  input_dtype: float
  input_tensor: nchw

It works, but if I follow the directions and do:

model:
  path: plus://9ef45d9bed92e71b72ca560013078e80

I get the above error. Docs issue? Bug?

[NickM-27]

If you're on 0.16.1 it should be working as expected, no one else has had this issue running that.

[NickM-27]

What does 9ef45d9bed92e71b72ca560013078e80.json file look like

[JasonSwindle]

> cat 9ef45d9bed92e71b72ca560013078e80.json

{"id": "9ef45d9bed92e71b72ca560013078e80", "name": "yolov9s", "isBaseModel": false, "trainDate": "2025-09-07T18:45:37.636Z", "baseModel": "2025.2", "type": "yolo-generic", "supportedDetectors": ["onnx", "rocm", "openvino"], "width": 320, "height": 320, "inputShape": "nchw", "inputDataType": "float", "pixelFormat": "rgb", "labelMap": {"0": "person", "1": "dhl", "2": "deer", "3": "dog", "4": "horse", "5": "package", "6": "waste_bin", "7": "an_post", "8": "bird", "9": "robot_lawnmower", "10": "raccoon", "11": "boat", "12": "fox", "13": "umbrella", "14": "bear", "15": "cat", "16": "license_plate", "17": "purolator", "18": "bbq_grill", "19": "cow", "20": "face", "21": "usps", "22": "ups", "23": "amazon", "24": "postnl", "25": "fedex", "26": "bicycle", "27": "squirrel", "28": "nzpost", "29": "car", "30": "motorcycle", "31": "postnord", "32": "gls", "33": "dpd", "34": "goat", "35": "rabbit", "36": "kangaroo", "37": "skunk", "38": "canada_post", "39": "royal_mail", "40": "school_bus"}, "verifiedCounts": {"retaining_wall": {"deer": 303, "images": 303, "package": 303, "fox": 303, "amazon": 303, "motorcycle": 303, "raccoon": 303, "face": 303, "squirrel": 303, "license_plate": 303, "fedex": 303, "car": 303, "person": 303, "cat": 303, "usps": 303, "bird": 303, "rabbit": 303, "ups": 303, "bear": 303, "dog": 303}, "driveway": {"deer": 421, "images": 421, "package": 421, "bicycle": 421, "dhl": 421, "fox": 421, "amazon": 421, "motorcycle": 421, "raccoon": 421, "face": 421, "squirrel": 421, "license_plate": 421, "fedex": 421, "car": 421, "person": 421, "cat": 421, "usps": 421, "bird": 421, "rabbit": 421, "ups": 421, "bear": 421, "dog": 421}}}

[hawkeye217]

Can you post the first portion of your logs on Frigate startup?

[JasonSwindle]

Oddly, after applying:

  model_type: yolo-generic
  input_dtype: float
  input_tensor: nchw

and removing it, I can use the model. I have no earthly idea what the issue was and sadly the logs rolled away. I'm going to close this out. BTW, the AI that emailed me on this was spot on.