Following the steps mentioned in the hailo tutorial notebooks. I have tried to convert a model from .onnx to .hef for hailo8L.
The model - stereonet is publicly provided to be run only on Hailo8 but i have Hailo8L hat + Raspberry Pi 5.
So after obtaining the onnx file i did quantization using a dataset which provided me with a quantized har file.
Final step is to do compilation which will give hef file.
But it is in this step i am stuck as it is taking too long ~4 hours and still going.
My issue is that does it takes this much time to do compilation or am i doing something wrong.
There are warnings in output as well about unconnected layers.
Yes, that behavior is normal — especially when running with higher optimization levels. Some models can take several hours or even over a day to complete optimization.
Let me know if you need any help with the compilation process!
Thanks for the information. I will try the compilation again.
I was wondering if there’s any way to speed up the process. Would using a machine with more RAM or CPU help? I noticed that the GPU wasn’t being utilized during the compilation.
Kartik Pathak
Director | Menthosa Solutions Pvt. Ltd. www.menthosa.com
Having around 32 GB of RAM and a more powerful CPU will always help speed up compilation. The GPU, however, is only used during the quantization/optimization stage—not during parsing or the core compilation itself.