Unoptimized implementation of "Intrinsic Image Decomposition Using Structure-Texture Separation and Surface Normals" in Matlab and C++ MEX.

You can try demo.m using included RGB & depth image.
Included input RGB and depth images are extracted from NYU Depth V2 dataset.
Algorithm requires texture-removed RGB image. Authors' code of regcovsmoothing and RGF are included.
You can use any structure-texture separation algorithm instead of it.

1. Original code is tested on Matlab 2015a 64bit, Windows 7.
2. For other platform (32bit Windows, Linux, MacOS), compile the C++ mex source codes in 'mex' directory.
3. To compile mex source codes, corresponding ann and opencv library is required (please refer the compile.m).

Junho Jeon ([email protected])

Cite our papers if you find this software useful.

1. Junho Jeon, Sunghyun Cho, Xin Tong, Seungyong Lee, "Intrinsic Image Decomposition using Structure-Texture Separation and Surface Normals", European Conference on Computer Vision (ECCV 2014), September 2014.