Perceptual-based CNN model for watercolor mixing prediction

Abstract

In the poster, we propose a model to predict the mixture of water-color pigments using convolutional neural networks (CNN). With a watercolor dataset, we train our model to minimize the loss function of sRGB differences. In metric of color difference ΔELab, our model achieves 88.7 % of data that ΔELab < 5 on the test set, which means the difference can not easily be detected by human eye. In addition, an interesting phenomenon is found; Even if the reflectance curve of the predicted color is not as smooth as the ground truth curve, the RGB color is still close to the ground truth.