Yes, you can run your FOMO model trained with Edge Impulse on the Hailo accelerator (Raspberry Pi AI Kit). However, there are some important considerations:
Model Format Compatibility:
Our Hailo Dataflow Compiler (DFC) supports TensorFlow and TFLite models, but only in FLOAT32 format. It does not directly support INT8 quantized models.
Steps for Deployment:
a. Export your model from Edge Impulse as TensorFlow Lite (FLOAT32).
b. Use the Hailo Dataflow Compiler to quantize and compile the model to HEF format. During this process, you’ll need to specify the calibration data for INT8 quantization.
In summary, while it’s possible to run your FOMO model on the Hailo accelerator, you need to start with a FLOAT32 model and allow our Hailo tools to handle the quantization process. This ensures compatibility and optimal performance on our hardware.
If you need any further assistance with this process, please don’t hesitate to ask. We’re here to help you successfully deploy your model on our platform.