Figure 2.

Schematic of the RSA approach used in all experiments. (A) For analyzing the internal representations of the networks, we extracted the activation patterns to the stimuli in our set from a selection of layers. Subsequently, we computed RDMs based on the correlation distances (1 - Pearson r) between all pairs of stimuli in one type of depiction and for each given layer separately. This yielded three RDMs (one for each type of depiction) in every selected layer. (B) In order to obtain a measure of similarity across types of depiction regarding the internal representation in the networks, we computed the Spearman rank correlation between the lower triangular values of RDMs of the different types of depiction in each layer separately. For the RDMs comparing representations of photos and drawings, this yields a “photo-to-drawing” similarity.