Skip to content

maoLB/modified_srsf_method

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Modified version of the dense Semi-Rigid Scene Flow (SRSF) V 1.0 method

This is a modified version of the optical flow generation method dense Semi-Rigid Scene Flow (SRSF) V 1.0. The whole path handling has been reworked in order to be conform to another pipeline, which uses this method for comparison purpsoses. All rights belong to the original authors mentioned in the original readme below.

Original Readme

Binaries for running dense Semi-Rigid Scene Flow (SRSF) V 1.0

Copyright (c) 2014 Julian Quiroga


Terms of use

This program is provided for research purposes only. Any commercial use is prohibited. If you are interested in a commercial use, please contact the copyright holder.

If you used this program in your research work, you should cite the following publication:

@inproceedings{quirogaECCV2014, year={2014}, booktitle={European Conference on Computer Vision}, title={Dense Semi-rigid Scene Flow Estimation from RGBD Images}, keywords={motion; scene flow; RGBD image}, author={Quiroga, Julian and Brox, Thomas and Devernay, Frederic and Crowley, James}, pages={567-582}, }

This program is distributed WITHOUT ANY WARRANTY.


Install

For compiling, go to the source folder and type make


Data

i) RGB images should be stored in the folder "Images", in png format and following the label shown in the examples. ii) Depth images should be stored in the folder Images, in png format, 16 bits and following the label shown in the examples (Depth in mm). iii) Camera matrix should be stored in the folder "settings". iv) The 3 components of the Scene Flow {SFx,SFy,SFz} and the Optical Flow {OFx,OFy} are stored in the xml file SFlow in CvMat format. The matrix Flag takes the value 255 if a valid depth measure is available and so that an estimation for this pixel is available.


Usage

a) To run SRSF with default parameters:

./semirigSF num1 num2 Sel

with

*num1: first image number *num2: last image number *Sel: Selection = 0 <- Non-Rigid, 1 <- Rigid, 2 <- Rigid + Non-Rigid

Example:

./semirigSF 10 11 0 -----/ Non-rigid Scene flow between frames 10 and 11

./semirigSF 10 11 1 -----/ Rigid Scene flow between frames 10 and 11

b) To run LGSF adjusting some parameters:

** NON-RIGID estimation

./semirigSF num1 num2 0 Npyr Nwarps W Step MaxZ Alfa

with

Npyr: Levels of the pyramid
Nwarps: Number of alternations between GN algorithm and TV solver at each level of the pyramid
W: size of the window -- 2W+1 x 2W+1
Step: the algorithm process every "Step + 1" pixel (to run faster)
MaxZ: maximum depth to be processed (cm)
Alfa: Regularization weight

Example:

./semirigSF 20 21 0 2 3 2 1 140 10

** RIGID estimation

./semirigSF num1 num2 1 Npyr Step MaxZ

with

Npyr: Levels of the pyramid
Step: the algorithm process every "Step + 1" pixel (to run faster)
MaxZ: maximum depth to be processed (cm)

Example:

./semirigSF 10 11 1 2 1 100

** RIGID plus NON-RIGID estimation

./semirigSF num1 num2 2 Npyr Nalter Nwarps W Step MaxZ Alfa

with

Npyr: Levels of the pyramid
Nalter: Number of alternations between RIGID and NON-RIGID estimation at each level of the pyramid
Nwarps: Number of alternations between GN algorithm and TV solver at each level of the pyramid
W: size of the window -- 2W+1 x 2W+1
Step: the algorithm process every "Step + 1" pixel (to run faster)
MaxZ: maximum depth to be processed (cm)
Alfa: Regularization weight

Example:

./semirigSF 20 21 2 2 3 1 2 1 140 10

Additional parameters can be set directly in mainSRSF.cpp


Bugs and Comments

Please report to Julian Quiroga ([email protected])

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • OpenEdge ABL 72.4%
  • C++ 26.7%
  • Makefile 0.9%