How to specify different models when using different hardware detectors?

[ptr727]

I am (trying to) test Hailo-8 and OpenVino Intel GPU for detectors.
If I do not specify a model, then openvino fails on startup, if I specify a model per docs then openvino is happy to start.

If I do specify a model for OpenVino, I assume it will not be compatible with Hailo that uses a different model?
How do I specify a model for OpenVino and a different model for Hailo8, when there is just one model section?

[inkeep-ai-assistant[bot]]

---

Frigate supports using multiple detectors simultaneously, but you're correct that different detector types require different model formats(1). Unfortunately, Frigate currently does not support specifying different models for different detector types when using multiple detectors in the same configuration(1).

When you define multiple detectors, they all share the same model configuration section(1). This means if you configure a model path for OpenVINO (like an .xml file), the Hailo detector will also try to use that same path and fail, since Hailo requires .hef files(1).

Current Limitation

The documentation explicitly states: "Multiple detectors can not be mixed for object detection (ex: OpenVINO and Coral EdgeTPU can not be used for object detection at the same time)"(1). While this example mentions OpenVINO and Coral, the same limitation applies to mixing OpenVINO with Hailo-8(1).

Workaround Options

Since you cannot run both detector types simultaneously for object detection, you'll need to:

1. Test them separately - Configure one detector at a time, restart Frigate between tests
2. Choose one primary detector - Based on your testing results, select either OpenVINO or Hailo-8 for your deployment

OpenVINO Configuration

For OpenVINO with Intel GPU, you would use(1):

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

(1)

Hailo-8 Configuration

For Hailo-8, you would use(1):

detectors:
  hailo:
    type: hailo8l
    device: PCIe

model:
  width: 320
  height: 320
  input_tensor: nhwc
  input_pixel_format: rgb
  input_dtype: int
  model_type: yolo-generic
  labelmap_path: /labelmap/coco-80.txt

(1)

Note that when no custom model path is provided for Hailo, the detector automatically downloads the default YOLOv6n model based on your hardware(1).

📚 Sources:

1. Supported Hardware

---

Have another question? Just tag @inkeep.