Converting fully custom models (not YOLO) to .hef

General
,
1

Hi everyone,

I’m currently working on a project where I need to convert my own custom model (written from scratch) into a .hef file for deployment on the Hailo-8. This model is not based on YOLO, nor is it about fine-tuning existing models with a custom dataset. The reason is simple: it must integrate into already running company processes, so using pre-made architectures isn’t viable.

While searching for guidance, I noticed most tutorials revolve around YOLO models or custom datasets but not about converting entirely custom architectures. Specifically, I’m trying to figure out if any of these resources apply to my case:

  1. Cytron Tutorial: Raspberry Pi AI Kit - ONNX to HEF Conversion
  2. TowardsAI: Custom Dataset with Hailo AI Hat, YOLO, Raspberry Pi 5 and Docker
  3. Google Colab: Model Zoo-based Conversion Tutorial

Out of these, which one (if any) is the correct approach for converting a completely custom model architecture to .hef? Or is there another recommended tutorial/documentation that specifically covers this?

Additionally, I’m trying to understand the actual role of the Hailo Model Zoo:

  • Is it only for pre-made models?
  • Can it also be used for custom datasets with pre-made models?
  • Or does it offer any value for converting fully custom architectures as well?

While searching, I also found this Hailo community post which seems to ask a very similar question to mine. However, it looks like the person responding ran into issues and didn’t manage to get far with it, so it remains unclear whether that approach would help in my situation.

Any official guidance, personal experiences, or clear documentation references would be greatly appreciated.

Thanks in advance for your help!