Awesome
Vacancy: A Voxel Carving implementation in C++
Vacancy is a Voxel Carving (a.k.a. Visual Hull or Shape from Silhouette) implementaion in C++. Inputs are 2D silhouettes (binary mask) of target objects, corresponding camera parameters (both intrinsic and extrinsic) and 3D bounding box to roughly specify the position of the objects. Output is the reconstructed 3D model. In addition to naive one, supports KinectFusion like robust TSDF (Truncated Signed Distance Function) fusion.
Algorithm Overview
<img src="https://raw.githubusercontent.com/wiki/unclearness/vacancy/images/how_it_works.gif" width="640">- Initialize 3D voxels according to the input bounding box
- Compute SDF (Signed Distance Function/Field) from silhouette in 2D
- Volmetrically fuse 2D SDF values into the 3D voxels by using camera parameters
- Extract explicit 3D mesh from implicit surface representation in the 3D voxels
Output mesh
Two mesh extraction methods are implemented: voxel and marching cubes. Marching cubes are much better in practice while voxel representation is suitable to understand how the algorithm works.
| voxel | marching cubes |
|---|---|
| <img src="https://raw.githubusercontent.com/wiki/unclearness/vacancy/images/bunny_voxel.png" width="320"> | <img src="https://raw.githubusercontent.com/wiki/unclearness/vacancy/images/bunny_marching_cubes.png" width="320"> |
Build
To build sample bunny executable, use cmake with CMakeLists.txt in the top directory.
You can integrate Vacancy to your own projects as static library by cmake add_subdirectory() command.
Dependencies
Mandatory
Optional (can be disabled by cmake)
- stb
https://github.com/nothings/stb
- Image I/O
- OpenMP (if supported by your compiler)
- Multi-thread accelaration