ColorChecker Classic 24

[Roger-Breton]

I experimented with my ColorChecher Classic 24 using root6x3 model and refIlluminant D50.

I got these results which I don't think are great :

# cxcmtrain.m -> Color correction training results: ================================================================= mse errors: 0.007053 (avg), 0.0004896 (med), 0.1306 (max, #19) ciede00 errors: 3.609 (avg), 2.725 (med), 12.82 (max, #19) ciedelab errors: 6.033 (avg), 5.581 (med), 19.66 (max, #19) ================================================================= # Color correction validation results: ================================================================= mse errors: 0.007053 (avg), 0.0004896 (med), 0.1306 (max, #19) ciede00 errors: 3.609 (avg), 2.725 (med), 12.82 (max, #19) ciedelab errors: 6.033 (avg), 5.581 (med), 19.66 (max, #19) =================================================================

[QiuJueqin]

There are mainly two possible reasons for a poor regression results:
Was your camera's spectral sensitivity function analogous to a linear mapping of CIE color match function?
Did you correct spatial non-uniformity of your both training and validation images before extracting patches' colors?
@Roger-Breton