mrfstereo
所属分类:图形图像处理
开发工具:C++
文件大小:352KB
下载次数:218
上传日期:2009-01-20 10:21:50
上 传 者:
emperorlu
说明: 马尔科夫随机向量场立体匹配。计算机视觉中的经典算法。
(Markov random vector field stereo matching. Classical computer vision algorithms.)
文件列表:
mrfstereo (0, 2007-11-15)
mrfstereo\Makefile (285, 2007-11-15)
mrfstereo\mrfstereo.cpp (16551, 2007-06-05)
mrfstereo\tsukuba (0, 2007-05-08)
mrfstereo\tsukuba\tsukuba-imL.png (175014, 2006-09-21)
mrfstereo\tsukuba\tsukuba-imR.png (174571, 2006-09-21)
mrfstereo\XGetopt.cpp (6235, 2006-09-22)
mrfstereo\XGetopt.h (805, 2006-09-22)
mrfstereo version 1.0
(c) Daniel Scharstein 10/4/2006
Stereo matcher front-end to MRF library
Requires imageLib and MRF
The three directories should be within one directory.
The MRF directory (e.g., MRF2.1) should be called 'MRF':
rename it ("mv MRF2.1 MRF") or create a symbolic link ("ln -s MRF2.1 MRF").
To compile:
$ cd MRF
$ make
$ cd ../imageLib
$ make
$ cd ../mrfstereo
$ make
To run:
$ ./mrfstereo
mrfstereo version 1.0
usage: ./mrfstereo [options] imL imR dispL
reads imL and imR (in png or pgm/ppm format)
runs MRF stereo
writes dispL (in png or pgm/ppm format), disparities corresponding to imL
options:
-n nD disparity levels, by default 16 (i.e. disparites 0..15)
-b use Birchfield/Tomasi costs
-s use squared differences (absolute differences by default)
-t trunc truncate differences to <= 'trunc'
-a MRFalg 0-ICM, 1-Expansion (default), 2-Swap, 3-TRWS, 4-BPS, 5-BPM, 9-all
-e smoothexp smoothness exponent, 1 (default) or 2, i.e. L1 or L2 norm
-m smoothmax maximum value of smoothness term (2 by default)
-l lambda weight of smoothness term (20 by default)
-g gradThresh intensity gradient cue threshold
-p gradPenalty if grad < gradThresh, multiply smoothness (2 by default)
-o outscale scale factor for disparities (full range by default)
-w write parameter settings to dispL.txt
-x write timings to dispL.csv
-q quiet (turn off debugging output)
Exit 1
$ ./mrfstereo tsukuba/tsukuba-im* tsukuba/tsukuba.png
Reading image tsukuba/tsukuba-imL.png
Reading image tsukuba/tsukuba-imR.png
Image size: 384 x 288, nDisp = 16
Data term: absolute differences
Smoothness term: L1 norm, truncated at 2, lambda=20
******* Running Expansion for up to 500 x 1 iterations
Energy = 6913378 (Ed=6913378, Es=0) at start
Energy = 1134339 (Ed=930659, Es=203680), 3.1 secs
Energy = 1128470 (Ed=927190, Es=201280), 5.5 secs
Energy = 1127957 (Ed=926877, Es=201080), 8.0 secs
Energy = 1127957 (Ed=926877, Es=201080), 10.5 secs
scaling disparities by 17
Writing image tsukuba/tsukuba.png
近期下载者:
相关文件:
收藏者: