[Detector Support]: [GPU] The tensor size is not equal to model

[khalidaboelmagd]

Describe the problem you are having

I'm getting below error when using OV detector

[GPU] The tensor size is not equal to model, can't set input tensor with index: 0, because model input (shape=[1,300,300,3]) and tensor (shape=[1,320,320,3]) are incompatible

I need to understand from where shape=1,320,320,3 coming from, which process is responsible for it.

Version

0.15-beta4

Frigate config file

mqtt:
  host: core-mosquitto
  user: mqttuser
  password: mqttuser
  topic_prefix: frigate
  client_id: frigate
  stats_interval: 60
  enabled: true

database:
  path: /config/frigate.db

detectors:
  ov:
    type: openvino
    device: GPU
    model:
      width: 300
      height: 300
      input_tensor: nhwc
      input_pixel_format: bgr
      path: /openvino-model/ssdlite_mobilenet_v2.xml
      labelmap_path: /openvino-model/coco_91cl_bkgr.txt

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

go2rtc:
  streams:
    camera-1:
      - rtsp://
    doorbell:
      - rtsp://

cameras:
  camera-1:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/camera-1
          roles:
            - record
            - detect
            - audio
    detect:
      fps: 5
    objects:
      track:
        - person
        - car
        - motorcycle
        - bicycle
        - truck
        - dog
        - cat
        - cycle
      filters:
        person:
          min_area: 800
          max_area: 25000
          threshold: 0.6
        car:
          min_area: 800
          max_area: 20000
          threshold: 0.65
        motorcycle:
          min_area: 500
          max_area: 18000
          threshold: 0.6
        bicycle:
          min_area: 400
          max_area: 15000
          threshold: 0.6
        truck:
          min_area: 1000
          max_area: 30000
          threshold: 0.65
        dog:
          min_area: 300
          max_area: 12000
          threshold: 0.6
        cat:
          min_area: 200
          max_area: 10000
          threshold: 0.6
        cycle:
          min_area: 400
          max_area: 15000
          threshold: 0.6
    record:
      enabled: true
      retain:
        days: 14
        mode: motion
      alerts:
        retain:
          days: 60
        pre_capture: 15
        post_capture: 20
      detections:
        retain:
          days: 30
    snapshots:
      enabled: true
      timestamp: true
      bounding_box: true
      retain:
        default: 90
    motion:
      threshold: 43
      contour_area: 35
      improve_contrast: true
      mask:
        - 0.995,0,0.723,0,0.73,0.039,0.747,0.147,0.755,0.191,0.765,0.269,0.761,0.33,0.771,0.388,0.793,0.472,0.805,0.513,0.803,0.555,0.804,0.593,0.814,0.634,0.867,0.678,0.905,0.702,0.997,0.726
        - 0,0.002,0.261,0,0.255,0.191,0,0.29
        - 0.717,0,0.748,0.148,0.253,0.184,0.264,0
    zones:
      kia:
        coordinates:
          0.636,0.372,0.633,0.331,0.623,0.297,0.613,0.281,0.6,0.273,0.58,0.27,0.565,0.275,0.552,0.295,0.547,0.31,0.546,0.345,0.548,0.379,0.567,0.395,0.595,0.406,0.628,0.404
        inertia: 3
        loitering_time: 0
      face:
        coordinates:
          0,0.322,0.1,0.312,0.394,0.382,0.647,0.41,0.781,0.472,0.817,0.993,0.005,0.994,0.002,0.413
        inertia: 3
        loitering_time: 0
        objects: person
    audio:
      enabled: true
    review:
      detections: {}
      alerts:
        required_zones: kia
  doorbell:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/doorbell
          roles:
            - detect
            - record
            - audio
    detect:
      fps: 5
    objects:
      track:
        - person
        - cat
        - dog
      filters:
        person:
          threshold: 0.6
        cat:
          threshold: 0.6
        dog:
          threshold: 0.6
    record:
      enabled: true
      retain:
        days: 14
        mode: motion
      alerts:
        retain:
          days: 60
        pre_capture: 15
        post_capture: 25
      detections:
        retain:
          days: 30
    snapshots:
      enabled: true
      timestamp: true
      bounding_box: true
      retain:
        default: 90
    motion:
      threshold: 30
      contour_area: 10
      improve_contrast: true
      mask:
        - 0.629,0,0.138,0,0.015,0,0.184,0.328,0.345,0.27
    zones:
      door:
        coordinates: 1,0,0.634,0.002,0.57,0.571,0.569,0.869,0.528,0.923,0.522,0.997,0.989,0.984
        inertia: 3
        loitering_time: 0
        objects:
          - person
    audio:
      enabled: true
    review:
      detections: {}
      alerts:
        required_zones: door

semantic_search:
  enabled: true
  reindex: false
  model_size: large
version: 0.15-0

docker-compose file or Docker CLI command

add on on homeassistant

Relevant Frigate log output

2025-01-02 05:32:19.556433110  [2025-01-02 07:32:19] frigate.detectors.plugins.openvino ERROR   : SSD model input doesn't match. Found [1,300,300,3].
2025-01-02 05:32:19.559389778  Process detector:ov:
2025-01-02 05:32:19.559393568  Traceback (most recent call last):
2025-01-02 05:32:19.559395017    File "/usr/lib/python3.9/multiprocessing/process.py", line 315, in _bootstrap
2025-01-02 05:32:19.559395943      self.run()
2025-01-02 05:32:19.559397064    File "/opt/frigate/frigate/util/process.py", line 41, in run_wrapper
2025-01-02 05:32:19.559397941      return run(*args, **kwargs)
2025-01-02 05:32:19.559398942    File "/usr/lib/python3.9/multiprocessing/process.py", line 108, in run
2025-01-02 05:32:19.559399849      self._target(*self._args, **self._kwargs)
2025-01-02 05:32:19.559400881    File "/opt/frigate/frigate/object_detection.py", line 145, in run_detector
2025-01-02 05:32:19.559401783      detections = object_detector.detect_raw(input_frame)
2025-01-02 05:32:19.559402766    File "/opt/frigate/frigate/object_detection.py", line 95, in detect_raw
2025-01-02 05:32:19.559403663      return self.detect_api.detect_raw(tensor_input=tensor_input)
2025-01-02 05:32:19.559404740    File "/opt/frigate/frigate/detectors/plugins/openvino.py", line 157, in detect_raw
2025-01-02 05:32:19.559405632      infer_request.infer(input_tensor)
2025-01-02 05:32:19.559406691    File "/usr/local/lib/python3.9/dist-packages/openvino/runtime/ie_api.py", line 132, in infer
2025-01-02 05:32:19.559407529      return OVDict(super().infer(_data_dispatch(
2025-01-02 05:32:19.559408459  RuntimeError: Exception from src/inference/src/cpp/infer_request.cpp:116:
2025-01-02 05:32:19.559409314  Exception from src/inference/src/cpp/infer_request.cpp:66:
2025-01-02 05:32:19.559410824  Check 'shape.compatible(ov::PartialShape(tensor->get_shape())) || tensor->get_shape() == ov::Shape {0} || port.get_partial_shape().is_dynamic()' failed at src/plugins/intel_gpu/src/plugin/sync_infer_request.cpp:144:
2025-01-02 05:32:19.559412212  [GPU] The tensor size is not equal to model, can't set input tensor with index: 0, because model input (shape=[1,300,300,3]) and tensor (shape=[1,320,320,3]) are incompatible

Install method

HassOS Addon

Object Detector

OpenVino

Screenshots of the Frigate UI's System metrics pages

None

Any other information that may be helpful

No response

[halabut69]

In your config.yaml your 'model' entry is indented. This might just be an issue with copying and pasting the text but it should be non-indented. Check the frigate docs on object detectors for an example.

[khalidaboelmagd]

I don't think so that is the issue as i tried to format the yml still getting the same error

[halabut69]

The error is occurring in the intel GPU plugin, as a troubleshooting step, try running openvino on your CPU (change your config to 'device: CPU').

[khalidaboelmagd]

I tried CPU but same issue,
Also I've check detectors/detector_config.py and it set to 320x320

[NickM-27]

The original answer is correct, your config is wrong because model should be its own property not a property of detector.

If you've already changed this you need to show your config

[khalidaboelmagd]

you were correct, thank you.