[Detector Support]: Unable to get YOLOv9 working #20535
Replies: 5 comments 21 replies
-
|
Based on your configuration and the logs provided, there are a few potential issues with your YOLOv9 setup. Let me walk through the key points: YOLOv9 Configuration IssuesYour configuration looks mostly correct for YOLOv9 with OpenVINO, but there are some important considerations (1): detectors:
ov:
type: openvino
device: GPU
model:
model_type: yolo-generic
width: 320 # <--- should match the imgsize set during model export
height: 320 # <--- should match the imgsize set during model export
input_tensor: nchw
input_dtype: float
path: /config/yolov9-t-320.onnx
labelmap_path: /labelmap/coco-80.txtMissing Detection Process LogsFrom your logs, I notice that Frigate is starting successfully, but I don't see any specific detector initialization logs that would indicate whether the YOLOv9 model is loading correctly. The logs show the detector process starting but don't show model compilation or loading details. Intel Arc A310 CompatibilityYour Intel Arc A310 should work well with OpenVINO. According to the performance data, Intel Arc A310 shows good inference times (2):
Troubleshooting Steps
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.txtLicense Plate Recognition IssueFor your LPR issue, the configuration shows The fact that your system metrics show the detector is running but with minimal activity suggests the model may not be loading properly or there might be an issue with the ONNX file itself. 📚 Sources: Have another question? Just tag @inkeep. |
Beta Was this translation helpful? Give feedback.
-
FINAL EDIT: got it to work! I had misunderstood that model was part of the detector configuration, moving it to root made it work 🥇I also cant get yolov9 to work. EDIT: I now see it seems to only be some remote detector, my error is from LocalDetector and there it should apply mappers... but seems to not work for some unknown reason...
tried both with CPU and GPU My detector config (nixos syntax) Errors I get: |
Beta Was this translation helpful? Give feedback.
-
|
The command I use to export the yolo model is this: I ran it in the terminal on my nas then copied the file output to the /config directory in frigate. I have also tried /config/model_cache but it doesn't make a difference. I also ran |
Beta Was this translation helpful? Give feedback.
-
|
What do you see in the debug view? Does the inference time change from 10 to something else? |
Beta Was this translation helpful? Give feedback.
-
|
Check the sidebar packages section on the frigate github https://github.com/users/blakeblackshear/packages/container/package/frigate |
Beta Was this translation helpful? Give feedback.
-
|
I manually updated openvino within the container, buy accessing the terminal. This fixed my issue. The version shipped with frigate is very old, I'd highly recommend updating it for others :) The versions shipped is |
Beta Was this translation helpful? Give feedback.
-
|
Yes, it's already updated for 0.17, as I said |
Beta Was this translation helpful? Give feedback.
-
|
@iambenmitchell what is the command you used to update openvino? thx |
Beta Was this translation helpful? Give feedback.
-
|
Openvino is a python package not an apt package. You use pip |
Beta Was this translation helpful? Give feedback.
-
|
What makes you think it's not working? You have a low inference speed which implies there are some detections happening. |
Beta Was this translation helpful? Give feedback.
-
|
It's not working because there are no detections or alerts showing up. Also in the main view on frigate, if I use another model it detects my parked car and shows the car icon on the camera preview. In debug mode, I can walk around by the camera and it is not detecting me. If I use a different model it does not see me either. |
Beta Was this translation helpful? Give feedback.
-
|
Changing the device to CPU from GPU fixes the issue. Is there something I am missing to get it to run on the GPU? |
Beta Was this translation helpful? Give feedback.
-
|
Check your use of "preset-intel-qsv-h265" ? as shown here in your config : ffmpeg: Eleswhere in your config you are using "preset-intel-qsv-h264"... In my case using consistent "preset-intel-qsv-h264" is the right course for my Intel implementation. |
Beta Was this translation helpful? Give feedback.
-
|
My cameras are H265 for the main feed, and the sub feed is h264. Right now I have the sub feed disabled, and I am using detect on the main feed. But there is no difference if I use the h264 sub feed for detect which is what I was doing earlier. I am just using the main feed at the moment test some stuff. |
Beta Was this translation helpful? Give feedback.
-
|
Ok. Gotcha |
Beta Was this translation helpful? Give feedback.
-
|
To also reiterate, it works fine with a different model. It just seems that if I use yolo-generic and provide the onyx file, detection doesn't work. This isn't an issue with the feeds as I have tested with other models, I can also see the feed fine in frigate, also under debug. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Describe the problem you are having
I have followed the instructions on the doc to download and export the yolo9 model for frigate.
I have placed it in the /config directory and have set the paths for it in the config. When frigate restarts, it seems detection isn't happening. If I use the ssdlite model, it works, as does YOLO-NAS. But when using yolo9, nothing is detected.
I am using openvino with an Arc A310
Version
0.16.2-4d58206
Frigate config file
docker-compose file or Docker CLI command
Relevant Frigate log output
Install method
Docker Compose
Object Detector
OpenVino
Screenshots of the Frigate UI's System metrics pages
Any other information that may be helpful
I am also unable to get the license plate recognition to work, no matter what model I select.
Beta Was this translation helpful? Give feedback.
All reactions