[Bug]: [GPU] Mismatch tensor and port type: f32 vs u8

[392media]

Describe the problem you are having

Running Frigate 0.14.0-beta3 on Unraid docker with an i5-6500T CPU. Config yolox-tiny as detector using openvino and Frigate crashes. Got the following error message in log:

Steps to reproduce

Start frigate with configuration below and it will crash.

Version

0.14.0-beta3

Frigate config file

detectors:
  ov:
    type: openvino
    device: GPU

model:
  width: 416
  height: 416
  input_tensor: nchw
  input_pixel_format: bgr
  model_type: yolox
  path: /config/yolox-tiny/FP16/yolox-tiny.xml
  labelmap_path: /labelmap/coco-80.txt

Relevant log output

2024-07-11 20:51:56.396659287  Process detector:ov:
2024-07-11 20:51:56.397518201  Traceback (most recent call last):
2024-07-11 20:51:56.397521231    File "/usr/lib/python3.9/multiprocessing/process.py", line 315, in _bootstrap
2024-07-11 20:51:56.397522843      self.run()
2024-07-11 20:51:56.397524600    File "/usr/lib/python3.9/multiprocessing/process.py", line 108, in run
2024-07-11 20:51:56.397528640      self._target(*self._args, **self._kwargs)
2024-07-11 20:51:56.397530370    File "/opt/frigate/frigate/object_detection.py", line 125, in run_detector
2024-07-11 20:51:56.397550784      detections = object_detector.detect_raw(input_frame)
2024-07-11 20:51:56.397553201    File "/opt/frigate/frigate/object_detection.py", line 75, in detect_raw
2024-07-11 20:51:56.397554831      return self.detect_api.detect_raw(tensor_input=tensor_input)
2024-07-11 20:51:56.397556483    File "/opt/frigate/frigate/detectors/plugins/openvino.py", line 155, in detect_raw
2024-07-11 20:51:56.397579695      infer_request.infer(input_tensor)
2024-07-11 20:51:56.397581689    File "/usr/local/lib/python3.9/dist-packages/openvino/runtime/ie_api.py", line 132, in infer
2024-07-11 20:51:56.397583411      return OVDict(super().infer(_data_dispatch(
2024-07-11 20:51:56.397593807  RuntimeError: Exception from src/inference/src/cpp/infer_request.cpp:116:
2024-07-11 20:51:56.397595340  Exception from src/inference/src/cpp/infer_request.cpp:66:
2024-07-11 20:51:56.397610505  Check 'port.get_element_type() == tensor->get_element_type()' failed at src/plugins/intel_gpu/src/plugin/sync_infer_request.cpp:144:
2024-07-11 20:51:56.397612065  [GPU] Mismatch tensor and port type: f32 vs u8
2024-07-11 20:51:56.397613368  
2024-07-11 20:51:56.397614632  

### Operating system

UNRAID

### Install method

Docker CLI

### Network connection

Wired

### Camera make and model

RTSP

### Any other information that may be helpful

_No response_

[NickM-27]

this may be because openvino 2024.1 is now used and these models generated are no longer compatible

[392media]

The model is converted from ONNX with the following tool:
OpenVINO Model Converter (ovc) 2024.2.0-15519-5c0f38f83f6-releases/2024/2

Is the tool's version too high? Should I use 2024.1 converter to convert the model?

[NickM-27]

probably worth a try to be sure

[krishne35]

any luck mate?