#StereoMatcher 1.0
This directory contains the source code to the two-frame dense stereo matching code developed by Daniel Scharstein and Richard Szeliski. Please see the associated Licence.txt and Copyright.h files for terms and conditions of use.
Modified by Vaibhav Desai to port DSP and other intensive SIMD type operations to GPU. This assumes an NVIDIA GPU using CUDA 6.0 or above.
The algorithms implemented in this code are described in our Technical Report,
D. Scharstein and R. Szeliski. A taxonomy and evaluation of dense
two-frame stereo correspondence algorithms. Technical Report
MSR-TR-2001-81, Microsoft Research, November 2001.
which can be found at
http://www.research.microsoft.com/scripts/pubs/view.asp?TR_ID=MSR-TR-2001-81
and which will also appear in a special issue of IJCV on stereo matching.
To compile the code, please use the Workspace or makefile.
If you wish to run the graph cut algorithm, you will need to follow the
instructions given in maxflow/README-maxflow.txt.
The program is driven by command line arguments and scripts. To run a particular scriptfile, just use
StereoMatch script scriptfile.txt
Some sample scripts that you can use to evaluate the output of your stereo algorithm or to run one of the matching algorithms are given on the Middlebury Stereo Web site,
http://www.middlebury.edu/stereo/code.html
You can also find some sample data sets at the same location.
There is no documentation provided beyond the comments embedded in the source code. The main parameters controlling the algorithm are in StereoParameters.h, and a subset of these is documented in more detail in the Techical Report cited above. If you don't understand what a parameter is doing, just leave it at its default value (it may control experimental code we are still working on).
You can obtain the latest version of this software by following the download link from the Middlebury Stereo Web page URL given above.
Originally developed by:
Daniel Scharstein and Richard Szeliski
December 6, 2001
Modified by:
Vaibhav Desai