Retrain Arface for Raspberry Pi 5 with AI HAT+ without GPU

Hello,

I’m currently working on a project where I need a face recognition model running on a Raspberry Pi with the Hailo AI HAT+. The model should be able to recognize individual faces in real time, and I need a way to retrain or configure it to identify specific people.

I found the ArcFace model from the Hailo Model Zoo, but I don’t have access to a GPU, so I can’t train it myself.

Is there any other way to adapt or retrain the model without a GPU? Or do you have alternative suggestions for achieving this with the Hailo AI HAT+?

Thanks in advance for your help!

The Hailo accelerators are design to run inference at the edge very efficiently not training.

For training you will need a GPU. It might be possible to do training for some models on a CPU.

@kaun
Can you please explain why you want to retrain? Is the current model not accurate enough? We tested this model and it is quite good. You can see a guide on adapting the model to recognize faces here: A Comprehensive Guide to Building a Face Recognition System

@shashi
I’m not sure if I need retraining at all, it was only mentioned here:
hailo_model_zoo/docs/RETRAIN_ON_CUSTOM_DATASET.rst at master · hailo-ai/hailo_model_zoo · GitHub
My goal is to have a model that can run on the Raspberry Pi 5 with AI HAT+ and which recognises the face of about 6 people in a photo and where you can also add new faces later on.

Do I understand correctly that your guide allows you to insert your own photos with the faces you want to be recognised? Can this run on the Raspberry Pi?

@kaun
yes, the guide shows how to start with a list of people you want to recognize and build a database of known faces (which can be extended by adding new faces). Yes, it runs on Raspberry Pi + Hailo.

Thank you @shashi ! I will try it