Speed up trt optimisation with lidar_centerpoint

[MarcFacerias]

Hello,

I'm working with docker and the lidar_centerpoint algorithm and I was wondering if there's some way to make an offline optimization of the onnx model. I have a very limited experience with this models, is it possible to serialize the optimized model so that it can be accessed later instead of rebuilding it ?

[MarcFacerias]

This feature is part of the lidar_centerpoint package, if someone needs to use it check the params build_only, head_engine_path and encoder_engine_path