18.02.2022

New article accepted for publication in the Journal of Vision

Title: "Deep neural networks for color classification and color constancy." by Alban Flachot, Arash Akbarinia, Heiko H. Schütt, Roland W. Fleming, Felix A. Wichmann, Karl R. Gegenfurtner

Abstract:
Color constancy is our ability to perceive constant colors across varying illuminations. Here, we trained deep neural networks to be color constant and evaluated their performance with varying cues. Inputs to the networks consisted of 2D images of simulated cone excitations derived from 3D-rendered scenes of 2115 different 3D-shapes, with spectral reflectances of 1600 different Munsell chips, illuminated under 278 different natural illuminations. The models were trained to classify the reflectance of the objects.
Testing was done with 4 new illuminations with equally spaced CIEL*a*b* chromaticities, 2 along the daylight locus and 2 orthogonal to it. High levels of color constancy were achieved with different deep neural networks, and constancy was higher along the daylight locus. When gradually removing cues from the scene, constancy decreased.
Both ResNets and classical ConvNets of varying degrees of complexity performed well. However, DeepCC, our simplest sequential convolutional network, represented colors along the 3 color dimensions of human color vision, while ResNets showed a more complex representation.

To see the whole article, please visit our publications page.