README for TensorRT 8.5.x / Jetpack 5.1.2 / Jetson Orin Nano (8GB Unified Memory)

step by step

1. build docker image

• docker build -t ur-image:v1 .

2. run dokcer container

• sudo docker run -it --rm --net=host --runtime nvidia -e DISPLAY=$DISPLAY -v /tmp/.X11-unix/:/tmp/.X11-unix
  -v ~/work:/workspace ur-image:v1

3. convert pt to onnx

• $ python3 onnx_convert.py

4. convert onnx to engine

• trtexec --onnx=model.onnx --saveEngine=model.engine --minShapes=input:1x3x64x64 --optShapes=input:1x3x256x256 --maxShapes=input:1x3x512x512 --fp16

5. numpy upgrade

• $ pip install --upgrade numpy

6. modify tensorRT init.py

•  

7. inference

• $ python3 engine_test.py

8. result

•  

•  

•