README.md
This code implements the simulation results reported in:
Please cite this paper if this code is used in, or to motivate, any publications.
REQUIREMENTS
This code was tested with MATLAB Version 9.3.0.713579 (R2017b) and MATLAB Version 9.4.0.813654 (R2018a) running on Ubuntu Linux version 16.04.
Dependencies:
The code uses some functions from the MATLAB Image Processing Toolbox.
The code uses the dataset and some functions from the Best-Buddies Template Matching code (version 1.0) available from: http://people.csail.mit.edu/talidekel/Best-Buddies%20Similarity.html The variable BBSPath in set_paths.m should be edited to point to the location where this code is installed.
Optional Dependencies:
Some functions use the Oxford VGG Affine Covariant Features Dataset which is available from: http://www.robots.ox.ac.uk/~vgg/research/affine/ To use these functions this data should be downloaded and the variable VGGPath in set_paths.m should be edited to point to the location of this dataset.
The code can also be used to test the DDIS and CoTM algorithms. To do so the code from https://github.com/roimehrez/DDIS and http://www.eng.tau.ac.il/~avidan/ should be installed and the variables DDISPath and CoTMPath in the file set_paths.m should be edited to point to the locations where this code is installed.
USAGE
To re-run the experiments described in the paper, do the following.
Fig. 1 can be reproduced using the command:
stats_correspondence_bbs_data(method,imagePair);
where method is set to the name of the algorithm to be tested (either 'DIM',
'ZNCC', 'BBS' or 'DDIS'), and imagePair is an integer between 1 and 105
specifying which image pair from the BBS dataset to use.
e.g. to reproduce the bottom-right panel, run: stats_correspondence_bbs_data('DIM',88);
Fig. 2 (and the corresponding values in Table 1) can be reproduced using the command:
stats_correspondence_bbs_data(method);
where method is set to the name of the algorithm to be tested (either 'DIM',
'ZNCC', 'BBS' or 'DDIS')
e.g. to reproduce the red curve, run: stats_correspondence_bbs_data('DIM');
Fig. 3 can be reproduced using the command:
expt_num_extra_templates_bbs_data;
To produce results with different methods of selecting the additional
templates requires editing of function template_matching_dim
Fig. 4 (and the corresponding values in Table 2) can be reproduced using the command:
stats_correspondence_vgg_data(method);
where method is set to the name of the algorithm to be tested (either 'DIM',
'ZNCC', 'BBS' or 'DDIS')
e.g. to reproduce the top-left panel, run: stats_correspondence_vgg_data('ZNCC');
To produce results with different sized templates, provide a value of 1, 2,
or 3 as the fourth input variable of the function (these numbers correspond
to the three columns of Fig. 4).
e.g. to reproduce the bottom-right panel, run: stats_correspondence_vgg_data('DIM',[],[],3);
Note that variable patchHalfLen defines the size of the patches before
downsampling.
Fig. 5 (and the corresponding values in Table 3) can be reproduced using the command:
stats_template_matching_vgg_data(method);
where method is set to the name of the algorithm to be tested (either 'DIM',
'ZNCC', 'BBS' or 'DDIS').
e.g. to reproduce the blue curve in the left panel run:
stats_template_matching_vgg_data('ZNCC');
To produce results with different sized templates, provide a value of 1, 2,
or 3 as the fourth input variable of the function (these numbers correspond
to the three columns of Fig. 5).
e.g. to reproduce the red curve in the right panel, run:
stats_template_matching_vgg_data('DIM',[],[],3);
Note that variable patchHalfLen defines the size of the patches before
downsampling.