Until now, only a limited number of algorithms on a limited number of data sets are available. If you have or know of any additions, please let me know.

You can browse the results per data set or per method. The first shows the results of all methods that are available for a particular data set. The second shows the results of one particular method on all data sets.

Methods:

—  Grey-Wozrld
—  White-Patch
—  Shades-of-Grey
—  General Grey-World
—  1^st-order Grey-Edge
—  2^nd-order Grey-Edge
—  Gamut Mapping
—  Spatial Correlations
—  Bayesian Color Constancy
—  Regression (SVR)
—  Inverse Intensity Chromaticity Space
—  Color Constancy using High-Level Visual Information
—  Color Constancy using Natural Image Statistics
—  CART-based Combination
—  Weighted Grey-Edge
—  Exemplar-Based Color Constancy
—  Various Univ. Zagreb methods
—  Using CNNs

Grey-World

Data set	Mean error	Median error	Trimean error	Max error
SFU Laboratory	9.8^o	7.0^o	7.6^o	37.3^o
SFU Grey-ball (Original)	7.9^o	7.0^o	7.1^o	48.1^o
SFU Grey-ball (Linear)	13.0^o	11.0^o	11.5^o	63.0^o
Color-checker (Original)	9.8^o	7.4^o	8.2^o	46.0^o
Color-checker (by Shi)	6.4^o	6.3^o	6.3^o	24.8^o

White-Patch

Data set	Mean error	Median error	Trimean error	Max error
SFU Laboratory	9.1^o	6.5^o	7.5^o	36.2^o
SFU Grey-ball (Original)	6.8^o	5.3^o	5.8^o	38.7^o
SFU Grey-ball (Linear)	12.7^o	10.5^o	11.3^o	46.5^o
Color-checker (Original)	8.1^o	6.0^o	6.4^o	36.3^o
Color-checker (by Shi)	7.6^o	5.7^o	6.4^o	40.6^o

Shades-of-Grey

The mentioned performance for every data set is with the optimal settings, which can differ per data set. For a full overview of the performance of different parameter settings, just download the results for every data set.

Data set	Mean error	Median error	Trimean error	Max error
SFU Laboratory	6.4^o	3.7^o	4.6^o	29.6^o
SFU Grey-ball (Original)	6.1^o	5.3^o	5.5^o	41.2^o
SFU Grey-ball (Linear)	11.6^o	9.7^o	10.2^o	58.1^o
Color-checker (Original)	7.0^o	5.3^o	5.6^o	36.6^o
Color-checker (by Shi)	4.9^o	4.0^o	4.2^o	22.4^o

General Grey-World

Data set	Mean error	Median error	Trimean error	Max error
SFU Laboratory	5.4^o	3.3^o	3.8^o	28.9^o
SFU Grey-ball (Original)	6.1^o	5.3^o	5.5^o	41.2^o
SFU Grey-ball (Linear)	11.6^o	9.7^o	10.2^o	58.1^o
Color-checker (Original)	7.0^o	5.3^o	5.6^o	36.6^o
Color-checker (by Shi)	4.7^o	3.5^o	3.8^o	22.0^o

1^st-order Grey-Edge

Data set	Mean error	Median error	Trimean error	Max error
SFU Laboratory	5.6^o	3.2^o	3.7^o	31.6^o
SFU Grey-ball (Original)	5.9^o	4.7^o	5.1^o	41.2^o
SFU Grey-ball (Linear)	10.6^o	8.8^o	9.2^o	58.4^o
Color-checker (Original)	7.0^o	5.2^o	5.5^o	36.3^o
Color-checker (by Shi)	5.3^o	4.5^o	4.7^o	26.4^o

2^nd-order Grey-Edge

Data set	Mean error	Median error	Trimean error	Max error
SFU Laboratory	5.2^o	2.7^o	3.3^o	26.7^o
SFU Grey-ball (Original)	6.1^o	4.9^o	5.3^o	41.7^o
SFU Grey-ball (Linear)	10.7^o	9.0^o	9.4^o	56.0^o
Color-checker (Original)	7.0^o	5.0^o	5.4^o	38.0^o
Color-checker (by Shi)	5.1^o	4.4^o	4.6^o	23.9^o

Gamut Mapping

Data set	Mean error	Median error	Trimean error	Max error
SFU Laboratory	3.6^o	2.1^o	2.4^o	27.1^o
SFU Grey-ball (Original)	6.9^o	5.8^o	6.1^o	41.9^o
SFU Grey-ball (Linear)	11.8^o	8.9^o	10.0^o	47.5^o
Color-checker (Original)	6.9^o	4.9^o	5.2^o	37.1^o
Color-checker (by Shi)	4.2^o	2.3^o	2.9^o	24.2^o

Spatial Correlations

The mentioned performance for every data set is with regularization. For a full overview of the performance with and without regularization, just download the results for every data set.

Data set	Mean error	Median error	Trimean error	Max error
SFU Laboratory — HeavyTailed-based	5.6^o	3.5^o	4.3^o	21.6^o
SFU Grey-ball (Linear) — Gaussian-based	12.7^o	10.8^o	11.5^o	41.2^o
— HeavyTailed-based	10.3^o	8.9^o	9.2^o	53.6^o
Color-checker (by Shi) — Gaussian-based	4.0^o	3.1^o	3.3^o	19.8^o
— HeavyTailed-based	3.6^o	3.0^o	3.0^o	21.6^o

The results of the Spatial Correlations denoted “Gaussian-based” are computed by Gijsenij et al. using an earlier implementation of the Spatial Correlations method. The results of Spatial Correlations denoted “HeavyTailed-based” is using the newest implementation and can be found in “Color Constancy with Spatio-spectral Statistics”, in IEEE TPAMI. More elaborate results by Chakrabarti et al. can be downloaded here. These results also contain slightly different results for the Grey-Edge-type methods.

Bayesian Color Constancy

Data set	Mean error	Median error	Trimean error	Max error
Color-checker (Original)	6.7^o	4.7^o	5.0^o	39.4^o
Color-checker (by Shi)	4.8^o	3.5^o	3.9^o	24.5^o

Regression (svr)

Data set	Mean error	Median error	Trimean error	Max error
SFU Grey-ball (Linear)	13.1^o	11.2^o	11.8^o	59.6^o
Color-checker (by Shi)	8.1^o	6.7^o	7.2^o	32.0^o

Inverse Intensity Chromaticity Space

Data set	Mean error	Median error	Trimean error	Max error
SFU Laboratory	15.5^o	8.2^o	10.7^o	80.9^o
SFU Grey-ball (Original)	6.6^o	5.6^o	5.8^o	76.2^o
SFU Grey-ball (Linear)	14.7^o	11.0^o	11.6^o	86.8^o
Color-checker (Original)	9.7^o	6.0^o	6.7^o	61.3^o
Color-checker (by Shi)	13.6^o	13.6^o	13.5^o	56.7^o

Color Constancy using High-Level Visual Information

The mentioned performance for every data set is the combination of bottom-up and top-down approaches. For a full overview of the performance of bottom-up, top-down and the combination of the two, just download the results for every data set.

Data set	Mean error	Median error	Trimean error	Max error
SFU Grey-ball (Linear)	9.7^o	7.7^o	8.2^o	60.0^o
Color-checker (Original)	6.4^o	4.5^o	5.0^o	33.6^o
Color-checker (by Shi)	3.5^o	2.5^o	2.6^o	25.2^o

Color Constancy using Natural Image Statistics

The mentioned performance for every data set is obtained using cross-validation on all considered methods and classifier-parameters.

Data set	Mean error	Median error	Trimean error	Max error
SFU Grey-ball (Original)	5.2^o	3.9^o	4.3^o	44.5^o
SFU Grey-ball (Linear)	9.9^o	7.7^o	8.3^o	56.1^o
Color-checker (Original)	6.1^o	4.5^o	4.9^o	36.8^o
Color-checker (by Shi)	4.2^o	3.1^o	3.5^o	26.2^o

CART-based Combination

The results denoted “alt. ground truth” are recomputed by Bianco et al. using an alternative ground truth, which can be downloaded together with the results.

Data set	Mean error	Median error	Trimean error	Max error
Color-checker (by Shi) — Selection	4.5^o	3.4^o	3.6^o	24.7^o
— Combination	3.9^o	2.9^o	3.3^o	22.3^o
Color-checker (by Shi; alt. ground truth) — Selection	4.2^o	3.2^o	3.4^o	22.7^o
— Combination	3.7^o	2.9^o	3.1^o	15.0^o

Weighted Grey-Edge

The results on SFU Grey-Ball (Linear) are calculated on a subset, for computational reasons. The subset is the same as introduced by Bianco et al. in “Improving color constancy using indoor-outdoor image classification”. This subset consists of a predefined train- and testset and can be found here.

Data set	Mean error	Median error	Trimean error	Max error
SFU Laboratory (p = 7, ? = 5, ? = 8)	5.6^o	2.4^o	2.9^o	43.8^o
SFU Grey-ball (Linear)(SUBSET!) (p = 1, ? = 1, ? = 10)	10.4^o	8.9^o	9.2^o	50.4^o
Color-checker (Original) (p = 2, ? = 1, ? = 20)	6.6^o	4.7^o	5.1^o	44.3^o

Exemplar-Based Color Constancy

Data set	Mean error	Median error	Trimean error	Max error
SFU Grey-ball (Original)	4.4^o	3.4^o	3.7^o	45.6^o
SFU Grey-ball (Linear)	8.0^o	6.5^o	6.8^o	53.6^o
Color-checker (by Shi)	2.9^o	2.3^o	2.4^o	19.4^o

Various Univ. Zagreb methods

Project page corresponding to these method can be found here.

Data set	Mean error	Median error	Trimean error	Max error
SFU Grey-ball (Original) — ColorCat	4.2^o	3.2^o	3.5^o	43.7^o
— SmartColorCat	4.6^o	3.5^o	3.8^o	46.2^o
— ColorDog using WP / GW	5.3^o	3.7^o	4.2^o	46.8^o
— ColorDog using ColorCat	4.5^o	2.9^o	3.5^o	46.2^o
— ColorDog using Sm.ColorCat	4.8^o	3.1^o	3.7^o	48.7^o
— ColorAnts	5.0^o	3.7^o	4.1^o	46.5^o
SFU Grey-ball (Linear) — ColorCat	8.7^o	7.1^o	7.4^o	54.4^o
— SmartColorCat	8.2^o	6.3^o	6.7^o	54.6^o
— ColorDog using WP / GW	10.3^o	7.3^o	8.2^o	53.6^o
— ColorDog using ColorCat	8.8^o	6.0^o	7.0^o	54.8^o
— ColorDog using Sm.ColorCat	8.5^o	5.6^o	6.6^o	56.2^o
— ColorAnts	8.9^o	6.5^o	7.1^o	57.6^o

Using CNNs

Preprint of the article describing this method can be found here.

Data set	Mean error	Median error	Trimean error	Max error
Color Checker (By Shi) — AlexNet+SVR	4.7^o	3.1^o	3.5^o	29.2^o
— Deep color constancy	2.6^o	2.0^o	2.1^o	14.8^o
Color Checker (by Shi; alt. ground truth) — AlexNet+SVR	4.9^o	3.1^o	3.6^o	29.1^o
— Deep color constancy	2.8^o	2.0^o	2.1^o	18.8^o

Color Constancy

Research Website on Illuminant Estimation

Results per Method (Recovery error)

Methods:

Grey-World

White-Patch

Shades-of-Grey

General Grey-World

1^st-order Grey-Edge

2^nd-order Grey-Edge

Gamut Mapping

Spatial Correlations

Bayesian Color Constancy

Regression (svr)

Inverse Intensity Chromaticity Space

Color Constancy using High-Level Visual Information

Color Constancy using Natural Image Statistics

CART-based Combination

Weighted Grey-Edge

Exemplar-Based Color Constancy

Various Univ. Zagreb methods

Using CNNs

Methods:

Grey-World

White-Patch

Shades-of-Grey

General Grey-World

1st-order Grey-Edge

2nd-order Grey-Edge

Gamut Mapping

Spatial Correlations

Bayesian Color Constancy

Regression (svr)

Inverse Intensity Chromaticity Space

Color Constancy using High-Level Visual Information

Color Constancy using Natural Image Statistics

CART-based Combination

Weighted Grey-Edge

Exemplar-Based Color Constancy

Various Univ. Zagreb methods

Using CNNs

1^st-order Grey-Edge

2^nd-order Grey-Edge