Rao and Ballard (1999) start their paper by stating that extra-classical receptive fields, which exhibit the phenomenon of end-stopping are difficult to explain. They show that in an idealized predictive coding model trained on natural images, the "neurons" at a higher level in the model have a receptive field that exhibits end-stopping; therefore, if neurons in the brain are implementing predictive coding, they would exhibit this phenomenon.
What I don't understand is this - why is it difficult to explain extra-classical receptive fields? After all, retinal ganglion neurons that are on-center will also exhibit "end-stopping" if a stimulus extends outside of the center, and the explanation for that receptive field is straightforward: the photoreceptors at the periphery hyperpolarize the ganglion cell, whereas those at the center depolarize it. Why can't a similar story be told about cortical neurons? Why was the predictive coding framework needed to explain end-stopping?