rpi-5-hailo-8l-deploy-model.md

3.2 Deploy Model on Raspberry Pi 5 AI kit with Hailo-8L accelerator

Prerequisits

• Ensure you have a Raspberry Pi 5 with the Hailo-8L AI kit installed. Refer to the official guide for setup instructions: How to Set Up Raspberry Pi 5 and Hailo-8L
• Familiarize yourself with the basic detection pipeline provided by Hailo: Hailo RPi5 Basic Pipelines

1. Copy the model

Open a terminal on your local computer and copy the yolov8s.hef file to the Raspberry Pi. Replace placeholders with your actual paths and username. The user on the Raspberry Pi needs write permissions to the destination path.

scp /path/to/local/yolov8s.hef USERNAME@RASPBERRYPI_IP:/home/hailo/hailo-rpi5-examples/resources/

2. Configure Labels (on Raspberry Pi):

Navigate to the resources folder on your Raspberry Pi. Create a new JSON file named yolov8s_labels.json and adjust the following settings:

• detection_threshold: This value determines the minimum confidence score for a detection to be considered valid (default 0.5).
• max_boxes: This defines the maximum number of detections to be displayed (default 200).
• labels: This is a list containing the actual class names for your model (replace with your specific classes). Save this JSON file under /home/hailo/hailo-rpi5-examples/resources/yolov8_labels.json. Example yolov8s_labels.json:

{
    "detection_threshold": 0.5,
    "max_boxes":200,
    "labels": [
      "Vespa_velutina",
      "Vespa_crabro",
      "Vespula_vulgaris"
    ]
}

3. Run Inference:

Navigate back to the hailo-rpi5-examples directory and source your Hailo virtual environment:

source setup_env.sh

Run your model (rename yolov8s.hef and yolov8_labels.json to a more descriptive name). This will utilize the Hailo-8L accelerator to perform object detection using your YOLOv8 model on the Raspberry Pi camera input.

python3 basic_pipelines/detection.py --hef-path resources/yolov8s.hef --input rpi --labels-json resources/yolov8_labels.json