ALOT texture classification dataset

The following classification dataset and problems are derived from the Amsterdam Library of Textures (ALOT) color image database made of 250 classes of textures. Each of the 3 tables below correspond to a given variation in acquisition conditions. For technical details about the image acquisition conditions, see ALOT website.

Image files are named according to the following pattern: <class>_<camera><illumination>, where:

• <class> ∈ {001..250}
• <camera> ∈ {c1..c4}
• <illumination> ∈ {l1..l5,l8,i}, where l1..l5 are 5 different illumination directions, l8 is a hemispherical illumination (5 lights turned on) at 3075K, and i likewise at 2175K.

Download ALOT images at half resolution (768×512 pixels, 4GB)

Each problem is described by 3 files according to the Outex description:

• classes.txt (identical for all problems) contains the number of classes, then the image, class number, and cost information (unused here) for each class
• test.txt contains the number of test images, then the image file name and (true) class number for each image
• train.txt contains the number of train images, then the image file name and class number for each image

Download all problems (see below for individual problem download)

Table 1: Classification datasets with illuminant variations (1 variation)

Table 2: Classification datasets with illumination direction variations (5 variations)

Table 3: Classification datasets with observation direction variations (3 variations)