pi-hailo-ml module

This module implements the rdk:service:vision API for the Raspberry Pi AI Kit and Raspberry Pi AI HAT+.

With this model, you can perform efficient AI inferencing, such as real-time image classification and object detection, while using minimum CPU resources compared to inferencing directly on the CPU.

Requirements

This module assumes you are using the AI Kit or AI HAT+ attached to the Raspberry Pi 5:

It also assumes you are using at least version Bookworm of the Raspberry Pi OS.

Configure your pi-hailo-ml Vision service

Navigate to the CONFIGURE tab of your machine in the Viam app. Add hipsterbrown:pi-hailo-ml to your machine.

Attributes

The following attributes are available for hipsterbrown:vision:hailo-rt vision service:

Name	Type	Required?	Default	Description
model	string *	Optional	'yolov8s_h8l'	Which pre-compiled ML model to use
labels_path	string	Optional	'assets/coco.txt'	Path to plain text file with the list of associated image labels for the model
default_minimum_confidence	number	Optional	0.55	Number between 0 and 1 as a minimum percentage of confidence for the returned outputs from the model

* This value must be one of the included pre-compiled models, see more below.

Model Selection:

While this module tries to select the best general model, feel free to experiment with the various included models to see if another works better for your use case.

You can learn more about the available models in the Hailo Model Zoo.

The following models can be used:

• yolov6n_h8
• yolov6n_h8l
• yolov8s_h8
• yolov8s_h8l
• yolox_s_leaky_h8l_rpi

Example configuration

{
    "model": "yolox_s_leaky_h8l_rpi"
}

Next steps

Compile other ML models you've trained or found from a model zoo like HuggingFace. https://github.com/hailo-ai/hailo-rpi5-examples?tab=readme-ov-file#hailo-dataflow-compiler-dfc

Guide coming soon.

Add support for custom model paths to use any supported models from the Hailo Model Zoo.

Troubleshooting

Mis-matching Hailo firmware and drivers

If you see an error related to the Hailo firmware (hailofw or hailo-dkms) version not matching the runtime drivers (hailort or python3-hailort), install the correct version of the software from the Hailo Developer Zone. You should search for "Vision Processors", selecting the following form options:

• "Archive" instead of "Latest releases"
• "HailoRT" in "Software Sub-Package"
• "ARM64" for "Architecture"
• "Linux" for "OS"
• Whichever Python version is on your device, most likely 3.11

Download the "HailoRT - PCIe driver Ubuntu package (deb)", "HailoRT - Python package (whl)", and "HailoRT - Ubuntu package (deb)" of the version you want, then copy them over to your device.

Install the .deb packages using apt:

sudo apt install ./path/to/package.deb

Install the Python .whl package using the system's pip command:

sudo python3 -m pip install ./path/to/package.whl --break-system-packages

The --break-system-package flag is used to override the recommendation about installing packages in virtual environments and replace the system's version of the pythonr-hailort package.

You may need to remove this module from your machine configuration, save, then re-add it to use the new system packages on your device.