Jaykishan Y. Patel

The paper trains a deep convolutional neural network to estimate surface reflectance and tests its susceptibility to lightness illusions, finding it performs well compared to classical models.

This study compares how deep neural networks and human observers achieve lightness constancy by estimating albedo and illumination in controlled conditions.

The paper quantifies human perception invariance to image contrast and content using a generative model, finding a factorial structure in discrimination thresholds.