Fig 8.

(a) Training stimuli for experiment 2: Verniers were always presented alone. Lines were presented randomly in groups of 2, 4, 6, or 8. Cuboids and scrambled cuboids were presented in groups of two facing each other. The distance between the shapes varied randomly between 1–6 pixels. (b) Testing stimuli for experiment 2: In the vernier-alone condition (left), we expected the network to perform well on the vernier discrimination task. In the flanker condition, we expected crowding for all chosen shape types. However, we expect that discrimination performance for the cuboid flankers recovers with increasing routing iterations.