My name is Yarden, and I am an electrical engineering student. As part of an academic project, we are working with the Hailo-8 AI accelerator and need to perform a one-time conversion of an ONNX model to a HEF file in order to run it on the device.
We have set up the development environment on a Raspberry Pi, but due to compatibility limitations with the hailo_dataflow_compiler wheel, we are unable to complete the conversion on our end.
I kindly ask if it would be possible to receive your assistance with this one-time conversion. The model is strictly for academic, non-commercial use as part of our final project.
Please let me know if I can share the ONNX file with you for this purpose.
Hey
What do you mean by pytorch checkpoint?
Currently we have onnx format and we need to convert to hef format (by compiling I suggest)
How can you help us? Can we send you the file and you give back the hef format?
We trained on Yolov11 and want to run this model on HAILO AI 8, that the reason why we need hef file.
To be honest, we don’t know much about what you talked about, please explain a bit more
Hi @yarden_pardo
We have a compiler that starts with the pytorch checkpoint and goes through the steps of converting it into onnx and then to hef file. When you train a model in ultralytics, the final result is a pytorch checkpoint (the .pt file) which is then translated to onnx. After you have the hef file, you need some code to run input images or videos on this hef file. This is where our PySDK comes into picture. Simplifying Edge AI Development with DeGirum PySDK and Hailo, User Guide 3: Simplifying Object Detection on a Hailo Device Using DeGirum PySDK. This is how we can help you: you provide us the .pt file and some images from training (which we use for calibration). We compile and give you a hef file along with a model json and labels file (see our user guide above) that you can then run on Hailo8. Hope this helps.