YoloV5在tensorRT上加速（Ubuntu）(C++)

文章目录

• ​​1.软件安装​​

• ​​1.1 opencv安装​​
• ​​1.2 Tensorrt安装​​

• ​​2.编译tensorrtx/yolov5​​
• ​​3. INT8量化​​

1.软件安装

默认已经安装好了cuda、cudnn
我的cuda为11.1，cudnn为适配的版本

1.1 opencv安装

​​https://github.com/opencv/opencv/releases​​

tar xvf opencv-3.4.4.tar.gz
cd opencv-3.4.4
mkdir build
cd build
cmake ..
make
sudo make install

1.2 Tensorrt安装

​​https://developer.nvidia.com/nvidia-tensorrt-7x-download​​

解压压缩包

tar xvf TensorRT-7.2.3.4.Ubuntu-18.04.x86_64-gnu.cuda-11.1.cudnn8.1.tar.gz

环境变量设置

vim ~/.bashrc

export TR_PATH=/home/zc/yp/lib/TensorRT-7.2.3.4
export PATH=$PATH:$TR_PATH/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$TR_PATH/lib
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$TR_PATH/targets/x86_64-linux-gnu/lib

source ~/.bashrc

cd TensorRT-7.x.x.x/python
pip install tensorrt-7.x.x.x-cp3x-none-linux_x86_64.whl

cd TensorRT-7.x.x.x/graphsurgeon
pip install graphsurgeon-0.4.1-py2.py3-none-any.whl

cd TensorRT-7.x.x.x/uff
pip install uff-0.7.5-py2.py3-none-any.whl

进入到tensorrt目录下，将下列文件夹复制到对于系统文件夹

sudo cp -r ./lib/* /usr/lib
sudo cp -r ./include/* /usr/include

安装pycuda

pip install pycuda

测试TensorRT

2.编译tensorrtx/yolov5

​​https://github.com/wang-xinyu/tensorrtx/tree/yolov5-v5.0/yolov5​​​
python版本下（未做tensorRT加速）

3. INT8量化

将coco_calib.zip解压到build目录下

cd build
make clean
cmake ..
make

序列化模型

./yolov5 -s yolov5s.wts yolov5s.engine s

测试

./yolov5 -d yolov5s.engine ../samples