GitHub - pmonasse/stereo-guided-filter: Disparity map estimation in a stereo pair of images with cost volume filtering

pmonasse / stereo-guided-filter Public

Notifications You must be signed in to change notification settings
Fork 9
Star 26

Disparity map estimation in a stereo pair of images with cost volume filtering

Notifications

Name		Name	Last commit message	Last commit date
Latest commit History 38 Commits
data		data
third_party		third_party
CMakeLists.txt		CMakeLists.txt
Changelog.txt		Changelog.txt
README.txt		README.txt
cmdLine.h		cmdLine.h
costVolume.cpp		costVolume.cpp
costVolume.h		costVolume.h
filters.cpp		filters.cpp
image.cpp		image.cpp
image.h		image.h
io_png.c		io_png.c
io_png.h		io_png.h
license.txt		license.txt
main.cpp		main.cpp
main_weights.cpp		main_weights.cpp
occlusion.cpp		occlusion.cpp
occlusion.h		occlusion.h

Repository files navigation

Stereo Disparity through Cost Aggregation with Guided Filter
Pauline Tan, <[email protected]>, CMLA, ENS Cachan, France
Pascal Monasse, <[email protected]>, LIGM, Univ. Paris-Est, France
Version 1.2 released on 2023/03/22
Future releases and updates:
https://github.com/pmonasse/stereo-guided-filter

- Requirements
CMake for the build system (http://www.cmake.org/)

- Build
cd stereo-guided-filter_1.0
mkdir Build && cd Build
cmake -D CMAKE_BUILD_TYPE:string=Release ..
make

- Run
Usage: ./stereoGuidedFilter [options] im1.png im2.png dmin dmax

Options (default values in parentheses)
Cost-volume filtering parameters:
    -R radius: radius of the guided filter (9)
    -A alpha: value of alpha (0.9)
    -E epsilon: regularization parameter (6.5025)
    -C tau1: max for color difference (7)
    -G tau2: max for gradient difference (2)

Occlusion detection:
    -o tolDiffDisp: tolerance for left-right disp. diff. (0)

Densification:
    -O sense: fill occlusion, sense='r':right,'l':left
    -r radius: radius of the weighted median filter (19)
    -c sigmac: value of sigma_color (25.5)
    -s sigmas: value of sigma_space (9)

    -a grayMin: value of gray for min disparity (255)
    -b grayMax: value of gray for max disparity (0)

The parameter 'sense' used in densification is the direction of camera motion:
    - from left to right (value 'r'), common for Middlebury pairs
    - from right to left (value 'l')

- Output image files
disparity.png: disparity map after cost-volume filtering
disparity_occlusion.png: after left-right check
disparity_occlusion_filled.png: simple densification
disparity_occlusion_filled_smoothed.png: final densification with median filter

- Test
./stereoGuidedFilter -O r ../data/tsukuba0.png ../data/tsukuba1.png -15 0
Compare resulting image files with those in folder data.