to this dataset
F. and Funt, B.
A Large Image Database for Color Constancy Research,"
Proceedings of the Imaging Science and Technology Eleventh Color Imaging
Conference, pp. 160-164, Scottsdale, Nov. 2003.
Link to this dataset.
datasets were gathered as part of Kobus Barnard's Ph.D. research at SFU. Appropriate
archival references for the data accompany each dataset.
Calibration Data for Sony DXC-930.
Synthetic Test Data.
Colour Constancy Test
Images Captured Using Sony DXC-930.
Images of Objects Under
Spectra of Fluorescent
Object Recognition Image Database (Old
Version - 1998)
Re-processed Version of the Gehler Color
Constancy Dataset of 568 Images
This dataset is based on work done by Dr. Lilong
Shi at Simon Fraser University with Dr. Brian Funt.
It involves processing the original RAW format images provided by Gehler et al. to generate a new set of high
quality linear images. New ground truth illumination measurements are
also provided for these images.
original dataset is described in Peter Gehler and
Carsten Rother and
Andrew Blake and Tom Minka and Toby Sharp, "Bayesian
Color Constancy Revisited", Proceedings of the IEEE Computer
Society Conference on Computer Vision and Pattern Recognition, 2008.
The original Gehler dataset is available from here.
Peter Gehler for making his images publically
available and for permission to post our new versions of them. The new SFU
Shi version is available for download here. Note that
the black level offset needs to be subtracted from the original images. A Matlab template for loading the images and removing the
offset is available here.
Manually Labeled Color Constancy Datasets
Link to this dataset.
Bing Li, Weihua Xiong, Weiming Hu, Brian Funt
May 15, 2012
A total of 1,913 images are included in the 3 data sets. We manually
labeled each of these images with its the 3D
stages and its indoor/outdoor classification. Following Nedovic
et al., the 15 typical 3D stages sky+bkg+grd, bkg+grd, sky+grd, grd, nodepth, grd+Tbkg(LR), grd+Tbkg(RL), Tbkg(LR), Tbkg(RL), tbl+Prs+bkg, 1sd+wall(LR), 1sd+wall(RL), corner,
corridor, and prs+bkg are used. There are
separate files in both Matlab (.mat) and plain
text (.txt) formats containing the file name from the original data set, a
0/1 flag indicating indoor/outdoor, and a numerical stage value.
The image labeling provided here has been used in a paper entitled,
“Evaluating Combinational Illumination-Estimation Methods on
Real-World Images,” by Bing Li, Weihua Xiong, Weiming Hu, Brian Funt. This paper is
still under review (as of May 2012) and is not yet publically available. If
you use the label sets we provide here, please contact one of the authors
for information on how to cite the database. E-mail addresses are:
Mismatching Hue Circle Transition Wavelengths
Full precision data for Table II of
“Logvinenko, A.D., Funt, B., and Godau, C., "Metamer
Mismatching," IEEE Trans. on Image Processing, Vol. 23, No. 1, pp.
34-43, Jan. 2014” is available here. The
data files consist of 20 rows of the 5 transition wavelengths defining a
reflectance spectrum. Note that transitions T1, ..., T5 are given in order.
In some rows the wavelengths are increasing, some decreasing. For the
increasing case, the reflectance is zero for all wavelengths less than T1,
one from T1 to T2, zero from T2 to T3, and so on. For the decreasing case,
the reflectance is zero for all wavelengths greater than T1, one from T1
down to T2, zero from T2 down to T3, and so forth.