dk-liang / pytorch2libtorch

A demo for using C++ to inference the pytorch model

pytorch2libtorch

A demo for using C++ to inference the pytorch model

Introduction

Althought pytorch is efficient and convenient for developer, which still having many hinderer to deploy on many devices. Here, we demonstrate how to use c++ with libtorch to inference pytorch model.

Environment

python >= 3.6
pytorch >= 1.3
OpenCV >= 4.0.0 (It means C++ version instead of python.)

Pytorch-python

• cd ./pytorch_code and put the pretrain model into ./model
• Run python pth2lib.py to generate traced script model. The generated model can be found in ./model, which is named 'model_transfer.pt'.

LibTorch-C++

Now, we use C++ to inference the transfered model.

• cp ./pytorch_code/model/model_transfer.pt ./C++_inference/model/

• cd ./C++_inference

• Download LibTorch from here.

• Unzip the libTorch and rename as libtorch

• mv libtorch ./lib/

• Mkdir build directory. mkdir build

• Compile the project. cd ./build , run cmake.. and make

• Test. ./main ./data/test.jpg

.