These texture feature vectors are extracted from the 128x128 pixel tiles of 20 large aerial images.  These tiles are found in http://vision.ece.ucsb.edu/datasets/Aerial128x128Tiles/.

The feature vectors have dimension 50.  They are the mean and standard deviations of 24 Gabors filters (4 scales and 6 orientations).  Also included is the mean and standard deviation of the non-filtered image.  The exact ordering of the elements is not known.

There are 20 feature files.  The features are stored as floats in binary format, one feature at a time.

The features were extracted on a non-Intel machine so will need to be byte-swapped at the float level if read on an intel machine.

There are raw and normalized versions in the corresponding directories.

The following lists the large aerial image names, the number of vertical tiles, the number of horizontal tiles and the number of feature vectors:

haan1_126	41	41	1681
haan1_127	41	41	1681
haan1_128	41	41	1681
haan1_129	42	41	1722
gsem6_105	42	42	1764
gsem6_107	42	42	1764
hbiu_141	42	42	1764
pwsb1_34	42	42	1764
c4950_sf19	41	41	1681
c4950_sf20	42	42	1764
haan1_32	42	42	1764
haan1_33	41	41	1681
haan1_34	41	41	1681
pwsb1_18	42	42	1764
pwsb1_20	42	42	1764
pwsb1_21	42	42	1764
haoi_88		42	42	1764
haoi_90		42	41	1722
haoi_91		42	42	1764
haoi_92		42	42	1764

total				34698