From pytorch to rknn


I am developing/training squeezenet SSD network using pytorch and would like to run on rk3399pro NPU.

I tried to read in the model ( using rknn api and able to build it but fail at init_runtime and cannot move further…(see link here)

Question: what would be the recommended flow from .pt to .rknn? do we need to convert .pt to ONNX and then to .rknn? could anyone share how to do it by giving a sample code?

Thank you very much for your help.