Raspberry Pi 5 and Hailo and YOLO Conversion

General
1

Hello, thank you for reading my question.
Of course, I have purchased the Hailo and am using it with the Raspberry Pi 5.
I have watched many official resource videos.
I have researched many issues and solutions shared by experienced engineers.
However, as someone with limited experience, I still have many unclear areas where I seek assistance.
I have many custom-trained models from previous projects: yolov4 (weights) and yolov8 (pt).
From the official information, I understand that I should:
Step 1: Convert the custom-trained models to ONNX,
Step 2: Compile the model using Hailo Model Zoo,
This will generate a HEF file that can be used on the Raspberry Pi 5 AI Kit.

How can I convert the models to ONNX and compile them using Model Zoo?
Since I already have the custom-trained models’ setup on my desktop, can I skip using the following environments:
HailoRT – PCIe driver Ubuntu package (deb)
Hailo AI Software Suite – Docker training
Or do I need to use these methods to perform the conversion?

Thank you.

2

Hi @kun2917,
You can use the Hailo Model Zoo for creating a HEF, but notice that if your custom models have different structures compared to what the Model Zoo contains, it might not work as efficiently. In this case I recommend to use the Dataflow Compiler tool for conversion.
You can follow the instructions on how to do convert a model using the Dataflow Compiler user guide in the Hailo website under Developer Zone → Documentation.
You can also run, while inside the Suite docker, the CLI command “hailo -h” and it will give you a short description of the Dataflow Compiler’s pipeline.

If the custom model is just trained on different classes, it shouldn’t be an issue using the Model Zoo tool. You can see instruction on how to do so using theHailo Model Zoo user guide in the Hailo website under Developer Zone → Documentation.

Regards,

1 Like