LibTorch Example of SemanticSegmentation

https://blog.kerrycho.com/Image-Segmentation-Libtorch/

Current Support CMake

• Windows
• Linux
• MacOS

windows
mkdir build & cd build
cmake ..

linux
bash build.sh

Example

#Receive data through commandline.
1. Open cmd.exe

2. training or inference
#example training 
D:\vision>SemanticSegmentation-Libtorch.exe train data_dir 

#example inference
D:\vision>SemanticSegmentation-Libtorch.exe inference image_path model_path 

DataSet

# Only Coco data sets are supported.

diretory_root
 cocodataset
  -annotations
   -instances_train2017.json
   -instances_tran2017.json
   -...
  -train2017
   - train images
  -val2017
   - val images

Requirements

• OpenCV 4.1.1
• Libtorch 1.4
• CUDA 10.2
• Visual studio 2019
• Windows 10

Install with Scripts

SemanticSegmentation-Libtorch
 - Scripts
    -install.bat
    
Run install.bat
It will download and install automatically.

Convert Python Model to C++

support Only-Resenet backbones
convert.py
Run Python Script

test