Ultralytics YOLO

---

Ultralytics YOLO supports a wide range of models, from early versions like YOLOv3 to the latest YOLO11.

Configuration​

yolo:
  object_detector:
    model_path: /detectors/models/yolo/my_model.pt
    cameras:
      viseron_camera1:
        fps: 1
        scan_on_motion_only: true
        log_all_objects: false
        labels:
          - label: dog
            confidence: 0.7
            trigger_event_recording: false
          - label: cat
            confidence: 0.8

Pre-trained models​

These steps should assist in locating models, configuring your container to access them, and configuring Viseron to use them.

Finding models​

Pre-trained YOLO models can be found online or you can train them yourself.

Examples of where to find pre-trained models:

• Ultralytics
• Roboflow
• Hugging Face

There are models for many different tasks, including object detection. If you are not sure if there is a problem with Viseron please confirm your Viseron environment with a stock YOLO model from Ultralytics. For example: yolov8n.pt

This component does not provide any training capabilities. See the Ultralytics training documentation for more information.

Where to place models​

Place your YOLO models in a directory of your choice.

There will be a later step to map the directory to the container. Therefore, choose a location supported by docker compose. If in doubt, do not use a SMB or NFS share.

Configuring Docker to make models available to Viseron​

The following docker-compose.yaml snippet will show how to map the directory above to the container:

    volumes:
      - {models path}:/detectors/models/yolo

This is the only change to docker-compose.yaml required for this component.

Configuring Viseron to use a model​

Modify the model_path setting in your Viseron config.yaml to point to one of the model(s) you installed. See the example above.

Only one model can be used at a time.

Image resizing​

Images inferenced by the component are resized by the underlying ultralytics package to match the model's input size.

There is no functionality to resize the image in the yolo component configuration before inferencing.

Labels​

When Viseron loads the model, it will print that model's labels to the log.

cd {location of Viseron docker-compose.yaml}
docker compose logs | grep "Labels"
viseron  | 2025-05-29 08:19:04.943 [INFO    ] [viseron.components.yolo.object_detector] - Labels: {0: 'bicycle', 1: 'bird', 2: 'bus', 3: 'car', 4: 'cat', 5: 'dog', 6: 'motorcycle', 7: 'person', 8: 'truck', 9: 'squirrel', 10: 'car-light', 11: 'rabbit', 12: 'fox', 13: 'opossum', 14: 'skunk', 15: 'racoon'}

Troubleshooting​

logger:
  logs:
    viseron.components.yolo: debug