Connect Retinanet DLA model with NMS

Description

I’m following https://github.com/NVIDIA/Deep-Learning-Accelerator-SW/tree/main/scripts/prepare_models in attempt to use DLA for object detection, in particular the “retinanet_resnext50.py”

After the running python3 retinanet_resnext50.py and getting the .onnx file, I observed that the NMS layers are no where to be found, so it is not a proper object detection model…

Can you help point me to some resources or provide a cookbook on how to make the conversion end-to-end starting from an onnx file → tensorRT engine that used DLA and also includes retinanet outputs (bboxes, labels, etc.)

Environment

Orin AGX