Frigate and Frigate+ models question/confused.

[sstepane]

Hi,

I would like to clarify which models are supported in Frigate or Frigate+?
For example, I would like to make a datased and train yolov9-c model, convert it to my hardware (rknn) and use it with Frigate.
Are some model kinds are available only in Frigate+?
It's a bit vague in docs - what exact model kinds supported with what hardware/frigate version.

[inkeep-ai-assistant[bot]]

Frigate (open-source) and Frigate+ have different model ecosystems:

Frigate (Open Source) - Bring Your Own Model

You can use your own custom-trained models with Frigate. For Rockchip (RKNN) specifically, the following model types are supported: (1)

YOLO-NAS:

• deci-fp16-yolonas_s, deci-fp16-yolonas_m, deci-fp16-yolonas_l

YOLOv9:

• frigate-fp16-yolov9-t, frigate-fp16-yolov9-s, frigate-fp16-yolov9-m, frigate-fp16-yolov9-c, frigate-fp16-yolov9-e

YOLOx:

• rock-i8-yolox_nano, rock-i8-yolox_tiny, rock-fp16-yolox_nano, rock-fp16-yolox_tiny

You can also provide your own .rknn model by converting from ONNX format using the rknn-toolkit2. (1)

Frigate+ - Subscription Models

Frigate+ offers pre-trained models specifically optimized for security cameras. Three model types are available: mobiledet, yolonas, and yolov9. (2)

For Rockchip NPU, the recommended detector type is rknn with model type yolov9. (2)

Your Use Case

For training your own YOLOv9-c model and converting to RKNN:

1. Train your model and export to ONNX format
2. Place the ONNX file in config/model_cache/rknn_cache/onnx
3. Create a config/conv2rknn.yaml configuration file
4. Run docker exec <frigate_container_id> python3 /opt/conv2rknn.py to convert (1)

Example config for YOLOv9:

model:
  path: frigate-fp16-yolov9-c
  model_type: yolo-generic
  width: 320
  height: 320
  input_tensor: nhwc
  labelmap_path: /labelmap/coco-80.txt

(1)

Frigate+ models are subscription-only and trained on community-submitted security camera images, while open-source Frigate supports custom models you train yourself. (2)

📚 Sources:

1. Supported Hardware
2. Models

---

Have another question? Just tag @inkeep.