GPU Marching Cubes

This repository contains a GPU-accelerated implementation of the Marching Cubes algorithm, designed to extract a triangle mesh of an isosurface from a three-dimensional scalar field (volume data). The implementation utilizes CUDA (>=11.6) to perform calculations on the GPU, offering significant performance advantages over CPU-based solutions.

Features

• High Performance: Utilizes the computational power of modern GPUs to accelerate the Marching Cubes algorithm.
• Customizable SDF: Supports custom Signed Distance Functions (SDF) for isosurface generation, allowing for diverse geometric shapes.
• Configurable Resolution: Easy configuration of the voxel grid resolution, origin, and width through the Args struct.

Prerequisites

• NVIDIA GPU with CUDA Compute Capability 6.0 or higher.
• CUDA Toolkit 11.6 or higher.
• C++ compiler compatible with the CUDA version used (C++17 and cuda_std_17 or higher).
• CMake(>=3.18) for building the project.

Configuration

Before compiling the project, you must configure the Marching Cubes parameters and the Signed Distance Function (SDF):

1. Modify Parameters in main.cpp

Set the grid parameters and isovalue in main.cpp:

struct Args{
    uint3 resolution = make_uint3(200, 200, 200);
    double3 gridOrigin = make_double3(-2, -2, -2);
    double3 gridWidth = make_double3(4, 4, 4);
    double isoVal = 1.0; // Isovalue of level set
};

2. Set the Signed Distance Function (SDF)

We provide two interfaces for sdf computation.

2.1 CPU Version:

Modify the computeSDF function in main.cpp to change the SDF computation:

double computeSDF(const double3 &pos) {
    // Example: A sphere function
    return pos.x * pos.x + pos.y * pos.y + pos.z * pos.z;
}

2.2 GPU Version:

For computations directly on the GPU, modify the computeSDF function in MarchingCubes.cu:

__device__ double MCKernel::computeSDF(double3 pos) {
    // Example: A sphere function
    return pos.x * pos.x + pos.y * pos.y + pos.z * pos.z;
}

Compilation

mkdir build
cd build
cmake ..
cmake --build . -j your-core-num

This will compile the project into an executable named main. If there are any errors during the compilation, ensure that CMakeLists.txt is correctly set up to find CUDA and compile the .cu files.

Usage

Run the executable with the output path for the generated mesh:

./main output-path.obj

This command will generate a triangle mesh based on the configured SDF and save it to the specified .obj file.

The example code generates a mesh for a sphere with a radius of 1.