Skip to content

Data Flow Compiler

Work in progress

These are rough notes, not a finished write-up.

In this article, let us go deep into how to get AI models (pre-trained), convert it into .hef (Hailo Executable File.) and use it with Hailo-8 HAT for RPi. I have avoided optimised implementations for making this article simple.

We don't use any GPU, but require a linux machine to run Hailo DFC (Data Flow Compiler).

To make this understandable, I have used image similarity search as an example through out this discussion.

Terminologies

Finding pre-trained AI models.

What are the options available to view the ONNX files?

Netron

Install netron

pip install netron
netron <model>.onnx

This servers the model on a web-server which you can open on any browser. If you are running on a headless server, redirect the port so that you can open. There are many ways, but I prefer the following way. * ssh to server with port forwarding.


Parsing with hailo parser

hailo parser onnx  cas_vit_m.onnx