Introduce

This project is run the model on jetson nano with tensorRT

Lab flower & yolov8 folder contain two part , one is training model on colab , another is inference on nano

homework direction contain TODO Part and need complete by your self

nano requirement

install pycuda

sudo su
bash script/pycuda_install.sh

install onnxruntime

bash script/onnxruntime_install.sh

exprot onnx => tensorRT engine

onnx2trt.py can export your onnx model to tensorRT engine

# example
sudo python3 onnx2trt.py --onnx {model_path}

trtexec tool (official)

# export to onnx
/usr/src/tensorrt/bin/trtexec --onnx={onnx_model} --saveEngine={engine_path}
# check model performance
/usr/src/tensorrt/bin/trtexec --loadEngine={engine_path}  --warmUp=5000

Classify

Lab1 : Flower Classifiy with TensorRT

Homework1 : Direction Classifiy

Naki Teacher Lesson : Audio Classifiy

Object Detection

Lab2 : Yolov8 with TensorRT

Homework2 : Transportation Detection

Yolov8 Inference Speed Benchmark on Nano with MAXN mode

	ONNX RUNTIME CPU	ONNX RUNTIME CUDA	TensorRT FP32	TensorRT FP16
320*320 FPS	4.60	23.4	33.8	36.5
640*640 FPS	1.26	8.56	11.6	15.5