[Support]: yolov9 model compatibility 0.16.4 to 0.17.0 - fails on start

[bjsassone]

Checklist

• I have updated to the latest available Frigate version.
• I have cleared the cache of my browser.
• I have tried a different browser to see if it is related to my browser.
• I have tried reproducing the issue in incognito mode to rule out problems with any third party extensions or plugins I have installed.
• I have asked the AI at https://docs.frigate.video about my issue.

Describe the problem you are having

2026-03-01 06:02:11.097554565 Process frigate.detector:ov:
2026-03-01 06:02:11.097560137 Traceback (most recent call last):
2026-03-01 06:02:11.097562868 File /usr/lib/python3.11/multiprocessing/process.py , line 314, in _bootstrap
2026-03-01 06:02:11.097564724 self.run()
2026-03-01 06:02:11.097566890 File /opt/frigate/frigate/object_detection/base.py , line 143, in run
2026-03-01 06:02:11.097569138 object_detector = LocalObjectDetector(detector_config=self.detector_config)
2026-03-01 06:02:11.097571465 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-03-01 06:02:11.097573685 File /opt/frigate/frigate/object_detection/base.py , line 62, in init
2026-03-01 06:02:11.097579997 self.detect_api = create_detector(detector_config)
2026-03-01 06:02:11.097581688 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-03-01 06:02:11.097605560 File /opt/frigate/frigate/detectors/init.py , line 18, in create_detector
2026-03-01 06:02:11.097607384 return api(detector_config)
2026-03-01 06:02:11.097609353 ^^^^^^^^^^^^^^^^^^^^
2026-03-01 06:02:11.097611487 File /opt/frigate/frigate/detectors/plugins/openvino.py , line 45, in init
2026-03-01 06:02:11.097643906 self.runner = OpenVINOModelRunner(
2026-03-01 06:02:11.097645952 ^^^^^^^^^^^^^^^^^^^^
2026-03-01 06:02:11.097648085 File /opt/frigate/frigate/detectors/detection_runners.py , line 316, in init
2026-03-01 06:02:11.097650235 self.compiled_model = self.ov_core.compile_model(
2026-03-01 06:02:11.097651921 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-03-01 06:02:11.097654104 File /usr/local/lib/python3.11/dist-packages/openvino/_ov_api.py , line 610, in compile_model
2026-03-01 06:02:11.097657232 super().compile_model(model, device_name, {} if config is None else config),
2026-03-01 06:02:11.097659143 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-03-01 06:02:11.097660942 RuntimeError: Exception from src/inference/src/cpp/core.cpp:126:
2026-03-01 06:02:11.097687580 Exception from src/inference/src/dev/plugin.cpp:58:
2026-03-01 06:02:11.097689712 Exception from src/frontends/onnx/frontend/src/core/graph_cache.cpp:27:
2026-03-01 06:02:11.097691951 graph_input_cast_0 node not found in graph cache

Steps to reproduce

Upgrade
Start
Look at log

Version

0.17.0

In which browser(s) are you experiencing the issue with?

No response

Frigate config file

model:
  model_type: yolo-generic
  path: /config/custom_models/yolov9s.onnx
  width: 320
  height: 320
  input_tensor: nchw
  input_dtype: float
  labelmap_path: /config/custom_models/yolov9s-labels.txt

docker-compose file or Docker CLI command

n/a

Relevant Frigate log output

above

Relevant go2rtc log output

n/a

Operating system

Home Assistant OS

Install method

Home Assistant Add-on

Network connection

Wired

Camera make and model

n/a

Screenshots of the Frigate UI's System metrics pages

n/a

Any other information that may be helpful

No response

[inkeep-ai-assistant[bot]]

The error you're encountering appears to be related to OpenVINO failing to compile your custom YOLO model. Based on your configuration, you're using a custom YOLOv9 model with OpenVINO.

Your configuration shows:

model:
  model_type: yolo-generic
  path: /config/custom_models/yolov9s.onnx
  width: 320
  height: 320
  input_tensor: nchw
  input_dtype: float
  labelmap_path: /config/custom_models/yolov9s-labels.txt

(1)

The error Exception from src/frontends/onnx/frontend/src/core/graph_cache.cpp:27 suggests OpenVINO is having trouble loading or compiling your ONNX model file.

Here are some troubleshooting steps:

1. Verify your detector configuration - Ensure you have explicitly defined your detector in your config. The default Frigate configuration uses a CPU detector, but you need to specify OpenVINO explicitly:

detectors:
  ov:
    type: openvino
    device: GPU  # or CPU

(1)

2. Check model compatibility - YOLOv9 models are supported by OpenVINO, but the ONNX export must be done correctly. The model configuration you're using appears correct for YOLO-generic models(1).

3. Verify the model file - The error suggests the ONNX model file itself may be corrupted or incompatible with the OpenVINO version in Frigate 0.17.0. Try re-exporting your YOLOv9 model to ONNX format.

4. Try CPU mode first - If you're using GPU mode, temporarily switch to CPU to isolate whether this is a GPU-specific issue:

detectors:
  ov:
    type: openvino
    device: CPU

(1)

5. Check for model path issues - Ensure the path /config/custom_models/yolov9s.onnx is accessible from within the container and the file exists.

If the issue persists after these steps, you may want to test with one of the default OpenVINO models to confirm OpenVINO itself is working, then troubleshoot your custom model separately.

📚 Sources:

1. Supported Hardware

---

Have another question? Just tag @inkeep.

[hawkeye217]

Glad you were able to find the issue. Feel free to let us know if there's anything else we can help with.

[bjsassone]

Any chance of getting back that horizontal space in the UI? :-)

[hawkeye217]

We refactored the Settings layout to support full UI configuration in an upcoming version. Are you really having that much difficulty clicking points to make your masks and zones? How often are you making new or modifying your masks and zones?

[bjsassone]

Seems like quite a bit recently fine tuning detection so I don't get alerts every time the cat jumps onto the window box or the mail man walks through the yard, but I hear you. But, it does help to have the real-estate (eg doing it on the phone is not worth the frustration.) On my desktop w/ 0.16.4, I've got ~1150x860 to work with which is roughly 5/8 of the horizontal layout at near full screen. I've already downgraded from 17, so I can't measure, but it was like 25% less, so.... I suppose YMMV.

[hawkeye217]

It shouldn't be too hard to set up a keyboard hotkey to collapse the sidebar. I'll look at that for a future version of Frigate.

[bjsassone]

For those that might come across the same issue when upgrading to 0.17.0 from 0.16.4. Found that a yolov9 onnx model that worked on 0.16.4 requires exporting with opset=12 to work with the latest OpenVINO used by 0.17.0.