I have a Luxonis POE camera in which I'm trying to run my own trained models. I trained an SSD-512 detector using the mmdetection library, and converted it to onnx format using the mmdeploy library. I got a non-fp16 static version from one of the python conversion files, and a dynamic fp16 version from another python conversion file, both from the mmdeploy library. When trying to convert the onnx files to blob format, I'm running into two types of issues. Anyone know how to fix these?

Hi @ashmitg

  1. Model input has to be static for the conversion to complete.

  2. Input has to be of type FP16 or U8 (with conversion parameters) for the model to compile.

Thanks,
Jaka