yanhaozhang / mnt_and_sdf_cuda

Calculate the 2D SDF according to the boundary pixel of a binary silhouette image.

Parallel Moore-Neighbour Tracing and 2D SDF

This code is part of our TRO submission. It is a CUDA implementation of our parallel Moore-Neighbour Tracing and the 2D SDF.

Introduction

Moore-Neighbour Tracing is a common method to extract the boundary pixels from a binary silhouette. Please refer to here for some more details. The traditional MNT is not run in a parallel process. Here, we improve the traditional MNT such that it is calculated in a different way, and run in a parallel process. The CUDA implementation of our method is released. We also release the code of calculating the 2D SDF from the extracted boundary pixels.

Some more details:

• MATLAB code to calculate 2D SDF calculate_sdf2D.m
• MATLAB code to calculate 2D normal vector LineNormals2D.m (based on here).
• CUDA code in ./src/:
  • main function link MATLAB main_sdf.cpp
  • main function calculate SDF from a binary silhouette main_sdf_cuda.cu, main_sdf_cuda.h
  • some useful vector operators vector_ops.hpp
  • Calculate the normal vector of boundary line_normal.cu
  • calculate Parallel Moore Neighbour Tracing feature_contour.cu, feature_contour_utils.hpp
  • calculate SDF and the gradient of SDF. sdf_contour.cu, sdf_config.h
  • Some other functions cuda_utils.h

Usage

This code has been tested on Ubuntu 16.04 with CUDA 9.0, and MATLAB 2018a.

• Compile: In MATLAB command window, run calculate_sdf2D.m to compile the CUDA code. Some modification may be needed.
• Run: an example on how to use the code and a comparison with the MATLAB code is in main_PMNT_and_SDF2D.m.

Contacts:

Yanhao Zhang: [email protected]. Any discussions or concerns are welcomed :)