/yolov7_trt

a quick solution for yolov7 tensorRT deployment

Primary LanguageC++MIT LicenseMIT

yolov7_trt

a quick solution for yolov7 tensorRT deployment

preparation

  • first make sure tensorRT is installed in your machine.
  • install torch & torchvision
  • install torch2trt and other requirements
git clone https://github.com/NVIDIA-AI-IOT/torch2trt.git
cd torch2trt
python3 setup.py install
cd ..
pip3 install -r requirements.txt

xxx.pt -> xxx.onnx -> xxx.engine(xxx.pt)

  • first copy file export_ONNX_for_TRT.py to the path of yolov7 project ( yolov7 main page )
python3 export_ONNX_for_TRT.py --weights yolov7.pt --img-size 640 --batch-size 1 --simplify --opset 10

and yolov7_640x640.onnx and yolov7_640x640.yaml will be generated, copy both of them to the path of this project

python3 onnx2trt.py --onnx yolov7_640x640.onnx --yaml yolov7_640x640.yaml --workspace 8 --fp16

and yolov7_640x640.engine and yolov7_640x640.pt will be generated for tensorRT c++ inference and python inference

run

python

modify some params in detect.py and then

python3 detect.py

c++

modify your lib path in cpp/CMakeLists.txt and then

cd cpp
mkdir build && cd build
cmake ..
make
cd ..

modify params in run.py and then

python3 run.py