A Toolbox to Visualize Dense Image Correspondences
Creators
- 1. Bosch Corporate Research
- 2. Heidelberg Collaboratory for Image Processing (HCI)
Description
Introduction
In cooperation with the Robert Bosch GmbH, the Heidelberg Collaboratory for Image Processing provides some tools for the visualization of dense correspondences as they are estimated by stereo and optical flow algorithms. The visualizations have been realized in MATLAB and C++ and can be downloaded for research purposes under GNU public license. Tools for outlier-visualization on the one hand and fine-scale correspondences visualization on the other hand are implemented. Using fixed thresholds visual comparison of different algorithms also on real world scenes are possible. Additionally, we provide color-legends that allow for a one-to-one mapping between colors and estimated correspondences. The choices for designing the tools are motivated in this document [pdf].
Methods
Color Maps
The use of fixed thresholds allows for inter-algorithm comparisons and also to provide fixed color-maps for disparity and optical flow.
Figure 1: The legend for stereo visualization shows that objects with large disparity are encoded in reddish colors while objects with zero disparity are encoded in blue.
Figure 2: The legend for optical visualization shows how direction is mapped to color and magnitude is mapped to saturation.
Stereo Visualizations
For stereo we implemented three methods of visualization.
Figure 3: The fixed scaling (left) allows for comparable visualization, but does not exploit the full color-space. Cyclic encoding (middle) highlights small variations of the correspondences but gives a poor overview. Stretching between the minimal and maximal disparity, i.e. between 1 and 64 in this case, (right) allows to exploit the color-space more thoroughly but looses interpretability with the color map.
These visualizations can be combined to an overlay image that allows to visualize object boundaries and correspondence discontinuities simultaneously.
Figure 4: Overlay of correspondence visualizations and reference images.
Optical Flow Example Visualizations
For optical flow we implemented three methods of visualization.
Figure 5: The fixed scaling (left) allows for comparable visualization, but does not exploit the full space of color and saturation. Cyclic encoding (middle) highlights small variations of the correspondences but gives a poor overview. Adjusting be the component mean of the flow field, i.e. by -8.60 and 2.96 in this case, (right) allows to exploit the color-space more thoroughly but looses interpretability with the color map.
These visualizations can be combined to an overlay image that allows to visualize object boundaries and correspondence discontinuities simultaneously. However, in this case variations in saturation can be due to variations in the magnitude of the flow or variations in the image.
Figure 6: Overlay of correspondence visualizations and reference images.
Technical info
Code (in correspondenceVisualization.zip)
We provide a package containing the C++ and MATLAB code.
This package includes examples for the application of the toolbox and for the gerenration of the color-maps.
When using this code please note the disclaimer within the software.
A MATLAB and a C++ version of our visualization-code is provided.
For motivation and ideas implemented in this code see the file correspondence-visualization.pdf provided with these files.
For easier application the visualization tool comes with one example for flow visualization and one example for stereo visualization.
In MATLAB these examples are script that can be executed from the MATLAB prompt.
The C++ examples can be build under
- Windows using VisualStudio2010 ( the corresponding sln-files are stored in the build folder).
- Linux using make ( the Makefile is stored in the build folder ).
The implementation uses CImg for easier handling of images and their display. The necessary file CImg.h is provided in the include-folder.
Notes
Contact
In case you have questions regarding the use of the code, please contact Daniel Kondermann.
License
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% This file is part of the HCI-Correspondence Estimation Benchmark.
%
% More information on this benchmark can be found under:
% http://hci.iwr.uni-heidelberg.de/Benchmarks/
%
% Copyright (C) 2011 <Sellent, Lauer>
%
% This program is free software: you can redistribute it and/or modify
% it under the terms of the GNU General Public License as published by
% the Free Software Foundation, either version 3 of the License, or
% (at your option) any later version.
%
% This program is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
% GNU General Public License for more details.
%
% You should have received a copy of the GNU General Public License
% along with this program. If not, see <http://www.gnu.org/licenses/>.
%
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Files
Visualization.pdf
Files
(5.3 MB)
| Name | Size | Download all |
|---|---|---|
|
md5:f000ef50eff0eba0b8886e369110c718
|
477.8 kB | Preview Download |
|
md5:2867e8bbc07e9a42981049bc364a8229
|
6.3 kB | Preview Download |
|
md5:fc3810c2acdf3ceb955973c67044b06a
|
159.6 kB | Preview Download |
|
md5:1208314b807c585c4dca3017969f076c
|
262.2 kB | Preview Download |
|
md5:b7af78edb02eb2e2cfe84dc01c6444a6
|
388.8 kB | Preview Download |
|
md5:daa3f223187a34cecac4caf038ebc6a5
|
297.5 kB | Preview Download |
|
md5:3e0786b668438ff6b3fb1df3aba2af5f
|
389.0 kB | Preview Download |
|
md5:ad6219087f6002b3c65f574198585f0b
|
3.4 MB | Preview Download |
Additional details
Related works
- Obsoletes
- Software: https://hci.iwr.uni-heidelberg.de/content/toolbox-visualize-dense-image-correspondences (URL)