bamboosdu / Yolov3_TensorRT_BYD

Real-time object tracker on Jetson NX using YOLO v3 and sort.

.
├── common
├── deep_sort
├── eval_yolo.py
├── LICENSE
├── Makefile
├── plugins
├── __pycache__
├── pytrt.cpp
├── pytrt.pxd
├── pytrt.pyx
├── README.md
├── result_3.avi
├── run_yolo.sh
├── setup.py
├── trtNet.cpp
├── trtNet.h
├── trt_yolo.py
├── trt_yolo_with_screen.py
├── utils
└── yolo

The deep_sort file contains sort&deep_sort realization.

The plugins contains files about yolo net.

The utils contains files for visualization, preprocessing,etc.

The yolo is the most important one, it explains how to convert darknet weight to tensorRT engin.

NX部署检测模型

步骤

1、安装依赖

install_protobuf-3.8.0.sh

bash install_protobuf-3.8.0.sh

Install pycuda

pip install pycuda==2019.1.1

Install onnx==1.4.1(确保是这个版本，不然会出错)

sudo pip3 install onnx==1.4.1

2、生成依赖

cd ${HOME}/project/tensorrt_demos/plugins
make

3、生成tensorRT，加速推理

cd ${HOME}/project/tensorrt_demos/yolo
bash darknet2onnx.sh
bash onnx2trt.sh

4、Run Yolo

python3 trt_yolo.py

Run on video and now we can switch between kalman filter mode to smooth the results.

python3 trt_yolo_with_screen.py --video /home/zq/Videos/20201022.flv -m  yolov3-416