The participants should submit one dissimilarity matrix file for per method run on the whole database. The file must be an ASCII file and the filename of the matrix file should be the first author's last name, followed by the name of the method and the appropriate extension (.txt). For example, if the first author's name is Joe Smith and the name of the method is "LFD", you would submit your file as "smith_LFD.txt".
The dissimilarity matrix represents the dissimilarity of all pairs of models in the database. For N models, the matrix is a sequence of N by N floating point numbers, where the number at position i*N+j represents the dissimilarity between models i and j. A value of zero indicates that the descriptors are identical, and larger values indicate greater dissimilarity between the 3D models. All dissimilarity values should be positive and can be arbitrarily large.
For the track "Large Scale Comprehensive 3D Retrieval", there are 8,987 models, so the participants should submit the file with a 8,987 by 8,987 dissimilarity matrix. The archive file (for example "smith_LFD.zip") should be sent via e-mail. If the participants have more than one method, they should submit an archive file per method containing the matrix file (the filename should indicate the method).
An example matrix file for the test database can be downloaded here: smith_LFD.txt
Example dissimilarity matrix format for a sample database with 5 objects:
Assume that the dissimilarity between every two objects can be found in the following table,
Then the content of the dissimilarity matrix file should be as follows:
0.000000 51.063496 39.640095 40.890335 39.601498
51.063496 0.000000 48.407860 46.646378 44.762428
39.640095 48.407860 0.000000 34.667614 32.640121
40.890335 46.646378 34.667614 0.000000 30.316898
39.601498 44.762428 32.640121 30.316898 0.000000