[Detector Support]: Halio8 nmo longer finds things like robot_lawnmower and waste_bin

[ljmerza]

Describe the problem you are having

I switch from coral to hailo8 and latest plus models but now im only detecting things like person, cat, dog. It doesn't detect waste_bin and robot_lawnmower at all like the coral did

Version

0.16.2-tensorrt

Frigate config file

back_yard_side_camera:
    ffmpeg:
      inputs:
        - path: 
            rtsp://admin:{FRIGATE_AMCREST_PASSWORD}@192.168.1.140:554/cam/realmonitor?channel=1&subtype=0
          roles:
            - record
        - path: 
            rtsp://admin:{FRIGATE_AMCREST_PASSWORD}@192.168.1.140:554/cam/realmonitor?channel=1&subtype=1
          roles:
            - detect
    detect:
      width: 704
      height: 480
      fps: 5
    objects:
      track:
        - person
        - dog
        - cat
        - bird
        - deer
        - rabbit
        - squirrel
        - waste_bin
        - face
        - fox
        - raccoon
        - robot_lawnmower
        - possum

detectors:
  hailo:
    type: hailo8l
    device: PCIe
    model:
      # yolov9s 640x640 hailo8
      path: plus://8a66079f4db4577a78d2f12c9cfc2814

docker-compose file or Docker CLI command

frigate:
    image: ghcr.io/blakeblackshear/frigate:0.16.2-tensorrt
    container_name: frigate
    privileged: true
    restart: unless-stopped
    shm_size: 1024mb
    env_file: ./env/frigate.env
    cap_add:
      - CAP_PERFMON
    devices:
      - /dev/dri:/dev/dri # For intel hwaccel, needs to be updated for your hardware
      - /dev/hailo0:/dev/hailo0 # Passes a USB Hailo, follow driver instructions here https://docs.hailo.ai/hailo-8-linux-getting-started-guide/getting-started.html#2-install-the-hailo-linux-drivers
      - /dev/dri/renderD128:/dev/dri/renderD128 # For intel hwaccel, needs to be updated for your hardware
    # - /dev/bus/usb:/dev/bus/usb # Passes the USB Coral, needs to be modified for other versions
    # - /dev/apex_0:/dev/apex_0 # Passes a PCIe Coral, follow driver instructions here https://coral.ai/docs/m2/get-started/#2a-on-linux
    volumes:
      - ./volumes/frigate/config:/config
      - /media/cubxi/nvr/frigate:/media/frigate
      - /etc/localtime:/etc/localtime:ro
      - type: tmpfs
        target: /tmp/cache
        tmpfs:
          size: 4000000000
    ports:
      - 5001:5000
    networks:
      camera_network:
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

Relevant Frigate log output

no logs it doesnt find anything

Install method

Home Assistant Add-on

Object Detector

Other

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response

[inkeep-ai-assistant[bot]]

The issue you're experiencing is related to the difference between the default Frigate models and Frigate+ models. When you switched from Coral to Hailo8 with Frigate+ models, the available object types changed.

Frigate+ models support a specific set of objects that differs from the default models (1). According to the documentation, the available label types in Frigate+ models include waste_bin and robot_lawnmower under the "Other" category (1):

• Other: package, waste_bin, bbq_grill, robot_lawnmower, umbrella (1)

Since you're using a Frigate+ model (indicated by your path: plus://8a66079f4db4577a78d2f12c9cfc2814 configuration), these objects should be supported. However, there are a few things to check:

1. Verify your model configuration: You're using a YOLOv9 model which is supported on Hailo devices (1). For Hailo8/Hailo8L hardware, yolov9 is the recommended model type (1).

2. Check your object tracking configuration: Make sure you have the objects properly configured in your objects.track list, which you do have correctly set up.

3. Consider threshold adjustments: The documentation notes that when switching to YOLOv9, you may need to adjust your thresholds for some objects (1).

4. Model training: Frigate+ models improve through fine-tuning with your specific images. The documentation suggests submitting diverse images and both true and false positives to improve model accuracy (1). You should provide at least 100 verified images per camera with varying conditions (1).

Since you were detecting these objects with Coral but not with Hailo8, this suggests the Frigate+ model may need more training data for your specific environment to properly detect waste_bin and robot_lawnmower objects.

📚 Sources:

1. Models

---

Have another question? Just tag @inkeep.

[ljmerza]

I used frigate plus models for coral and it worked fine

[NickM-27]

Your config is invalid

detectors:
  hailo:
    type: hailo8l
    device: PCIe
    model:
      # yolov9s 640x640 hailo8
      path: plus://8a66079f4db4577a78d2f12c9cfc2814

needs to be

detectors:
  hailo:
    type: hailo8l
    device: PCIe

model:
  # yolov9s 640x640 hailo8
  path: plus://8a66079f4db4577a78d2f12c9cfc2814