This description shows how the test and training data is stored. The functions D_processTrainingData and D_processTestData compute the strings of a training and test set. Also the alignement matrix of the training strings against themselves and against the test data are computed.
The resulting data is stored as follows:



Strings of Training Data

These are stored in a SA_VectorArray using the public function save of this class. It is stored under the name 'TrainingDataStrings'. They can be loaded with matlab using the public function load.


Strings of Test Data

These are stored in a SA_VectorArray using the public function save of this class. It is stored under the name 'TestDataStrings'. They can be loaded with matlab using the public function load.


Alignment Matrix of Training Data with itself

The matrix is stored as ascii information into the file 'TrainingData_AlignmentMatrix'. The columns are separated by the '&' character and rows are separated with the newline character.


Alignment Matrix of Training Data with Test Data

The matrix is stored as ascii information into the file 'TestData_AlignmentMatrix'. The columns are separated by the '&' character and rows are separated with the newline character.


Information to reconstruct the images for matrix entries

To reconstruct which image was used to compute an entry of the alignment matrix, the matrix stored in 'TestData_IndexList' or 'TrainingData_IndexList' respectively can be used. The columns in these files are also separated by '&' and rows by the newline character.

The matrix stored in 'TrainingData_IndexList' is of size [M x 4] with M beeing the total number of strings extracted from the training data. Each of these 4 vectors [c, f, i, l] contains the following information about the string used to compute the corresponding entry of the training alignment matrix:

[c, f, i, l]:
c = category. 1 = anchor, 2 = crab, 3 = stopsign.
f = folder. This is the number of the folder which was used.
i = image. This is the number of the segmented image from which the boundaries were extracted. The number corresponds to the order in the final_results.dat file.
l = length. The length of the string wich was used to compute the alignment.

The information about the test data strings is stored in the same manner into the file 'TestData_IndexList'.




Copyright © 2007 Thomas Oskam, ETH Zürich. All rights reserved.