Fig. 4. Factors of objects’ task demands in inferring objects’ real-world size.

a Task demands. Three DCNNs with the same architecture (i.e., AlexNet) were trained to classify objects at different levels of categorization. DCNNs were trained to categorize a car as cab (basic level), conveyance (superordinate level), or artifact (coarse level). Note that the DCNN for the basic-level categorization is the same as the one used in the previous experiments. b The RSM of Conv4’s responses of the AlexNets with different task demands. From left to right: basic level, superordinate level, and coarse level. c The second axis of the object space also specifically encoded the real-world size with the mapping function of the common logarithm. S small objects, B big objects.