The Prague Texture Segmentation Datagenerator and Benchmark - Benchmark Data
optimized
for

Colour, grayscale, BTF or hyperspectral (ALI) benchmarks are generated upon request in three quantities (normal = base_size, large = 4 x base_size, huge = 9 x base_size test mosaics). The benchmark archive either in the compressed tar or zip formats contains images in PNG format. For each texture mosaic there are also the corresponding ground truth and mask images.
Colour Colour benchmark  (base_size=20)
Grayscale Grayscale benchmark  (base_size=20)
Colour (Rot. Inv.) Colour benchmark - rotation invariant  (base_size=20)
Colour (Scale Inv.) Colour benchmark - scale invariant  (base_size=20)
Colour (Rot.&Scale Inv.) Colour benchmark - rotation&scale invariant  (base_size=20)
Colour (Illum. Inv.) Colour benchmark - illumination invariant  (base_size=20)
Colour (Geom. Inv.) [exp.] Colour benchmark [experimental] - geometry invariant  (base_size=20)
ALI ALI benchmark - satellite multispectral image mosaics  (base_size=10)
ALI-rgb ALI-rgb benchmark - satellite rgb image mosaics  (base_size=10)
Colour (Rot.&Illum. Inv.) Colour benchmark - rotation&illumination invariant  (base_size=20)
BTF (var.L+fix.V) [exp.] BTF benchmark [experimental] - variable light, fixed view  (base_size=10)
BTF (fix.L+var.V) [exp.] BTF benchmark [experimental] - fixed light, variable view  (base_size=10)
BTF (var.L+var.V) [exp.] BTF benchmark [experimental] - variable light, variable view  (base_size=10)
GeoEye [exp.] GeoEye benchmark [experimental] - satellite rgb image mosaics  (base_size=10)
BTF wood [exp.] BTF wood benchmark [experimental] - real wood BTF textures  (base_size=10)
dataset size: normal (1x)  large (4x)  huge (9x)
classification: supervised (larger file, requires additional time to complete)
degradation type (noise/smoothing):      degrad. level (G. or S.&p. noise / G. blur / Median):
(Gaussian noise ... SNR;   Salt&pepper noise ... probability;   Gaussian blur ... sigma;   Median filter ... window half-size)
re-type:   here