Figure 2

Population receptive field mapping. (a) Schematic overview of the adapted pRF mapping method used for neural networks. (b) Retinotopic maps of convolutional layers. Each pixel in the maps reflects a population unit in the corresponding convolutional layer. (c) Convolutional units per level of eccentricity for the V1 layer, including the best fitting linear and exponential decay functions. (d) Receptive field size as a function of eccentricity for V1, V2, and V4 layers. Included in gray are the best linear fits for each layer.