The relationship of lightness illusions uncovered by individual differences and its advantage in model evaluation

Computational models that explain lightness/brightness illusions have been proposed. These models have been assessed using a simplistic criterion: the number of illusions each model can correctly predict from the test set. This simple method of evaluation assumes that each illusion is independent; however, since the independence and similarity among lightness illusions have not been well established, potential interdependencies among the illusions in the test set could distort the evaluation of models. Moreover, evaluating models with a single value obscures where the model’s strengths and weaknesses lie. We collected the magnitudes of various lightness illusions through two online experiments and applied exploratory factor analyses. Both experiments identified some underlying factors in these illusions, suggesting that they can be classified into a few distinct groups. Experiment 1 identified three common factors; assimilation, contrast, and White’s effect. Experiment 2, with a different illusion set, identified two factors–assimilation and contrast. We then examined three well-known models that are based on early-visual processes, using the outcomes of the experiments. The examination of these models revealed biases in the models towards specific factors or sets of illusions, which suggested their limitations. This study clarified that correlations of illusion magnitudes provide valuable insights into both illusions and models and highlighted the need to assess models based on their ability to account for underlying factors rather than individual illusions.