Awesome

Cross-Scale Cost Aggregation for Stereo Matching (CVPR 2014)

Compilation

Windows

The code is a Visual Studio 2010 project on Windows x64 platform. To build the project, you need to configure OpenCV on your own PC. (version 2.4.6, however, other versions are acceptable by modifying CommFunc.h).

Other Platforms

The code requires no platform-dependent libraries. Thus, it is easy to compile it on other platforms with OpenCV.

Usage

Run the program with the following paramters: Usage: [CC_METHOD] [CA_METHOD] [PP_METHOD] [C_ALPHA] [lImg] [rImg] [lDis] [maxDis] [disSc]

• [CC_METHOD] -- cost computation methods, currently support:
  • GRD -- intensity + gradient
  • CEN -- Census Transform
  • CG -- Census + gradient
• [CA_METHOD] -- cost aggregation methods, currently support:
  • GF -- guided image filter
  • BF -- bilateral filter
  • BOX -- box filter
  • NL -- non-local cost aggregation
  • ST -- segment-tree cost aggregation
• [PP_METHOD] -- post processing methods, currently support:
  • WM -- weighted median filtering
  • SG -- segment based (experimental)
• [C_ALPHA] -- regularization paramter, i.e. $\lambda$ in the paper.
• [lImg] -- input left color image file name. (all formats supported by OpenCV)
• [rImg] -- input right color image file name.
• [lDis] -- output left disparity map file name.
• [maxDis] -- maximum disparity range, e.g. 60 for Middlebury and 256 for KITTI dataets.
• [disSc] -- scale disparity, e.g. 4 for Middlebury and 1 for KITTI datasets.

Hint: to enable post-processing, you must uncomment // #define COMPUTE_RIGHT in CommFunc.h to allow computing right disparity map.

Citation

Citation is very important for researchers. If you find this code useful, please cite:

@inproceedings{CrossScaleStereo,
        author    = {Kang Zhang and Yuqiang Fang  and Dongbo Min and Lifeng Sun and Shiqiang Yang  and Shuicheng Yan and Qi Tian},
        title     = {Cross-Scale Cost Aggregation for Stereo Matching},
        booktitle = {CVPR},
        year     = {2014}
}

Since some cost aggregation methods (GF, NL, ST) are built uppon other papers' code, you also need to cite corresponding papers as listed below.

Reference

<a name="CT">[CT]</a>: R. Zabih and J. Woodfill. Non-parametric local transforms for computing visual correspondence. In ECCV, 1994

<a name="GF">[GF]</a>: C. Rhemann, A. Hosni, M. Bleyer, C. Rother, and M. Gelautz. Fast cost-volume filtering for visual correspondence and beyond. In CVPR, 2011

<a name="ST">[ST]</a>: X. Mei, X. Sun, W. Dong, H. Wang, and X. Zhang. Segment-tree based cost aggregation for stereo matching. In CVPR, 2013

<a name="BF">[BF]</a>: K.-J. Yoon and I. S. Kweon. Adaptive support-weight approach for correspondence search. TPAMI, 2006

<a name="NL">[NL]</a>: Q. Yang. A non-local cost aggregation method for stereo matching. In CVPR, 2012