The case of curved surfaces is easily handled by the model, as suggested in Figure 13 (A), and 14 (A) & (B) etc. [same figure numbers in the new draft] because the coplanarity interaction is not rigidly coplanar, but has a certain orientational tolerance, which means that the perceived surface (or dynamic sheet of active units in the volumetric representation) is free to bend and curve through smooth arcs that are locally near-coplanar. The curved surface is fundamental to a soap bubble, which serves as the guiding inspiration for this model.

As for surface intersections of other than 90 degrees, this is indeed an interesting problem, and one that must eventually be addressed. However in this paper I have taken on the problems of surface, corner, and occlusion completion by reification in an isomorphic perceptual model. That is already enough for one paper. If this paper does not address all problems in vision, that does not invalidate the issues which are eaddressed. After all, Marr and Biederman themselves do not address this issue, nor is there much mention of it anywhere in the modeling literature. The reviewer seems to suggest that an unconventional model must solve the entire vision problem in order to be given serious consideration. He should acknowledge however that the alternative conventional approach also fails to address these issues. It is sufficient therefore for a model to simply explain more phenomena than the alternative models.

As it happens, I actually do have a solution to the other-than-90 degree problem, which derrives from my thesis work (Lehar 1994), in which I showed that a large number of illusory groupings including both collinear and non-collinear grouping percepts through a variety of vertex types, can be explained by a general model involving harmonic resonances, or standing wave patterns in the neural substrate. That model generalizes nicely into 3-D too, but I have had difficulty publishing this paper despite having a very specific computational mechanism (and computer simulations), not for lack of solid supporting evidence, but because I dare to propose an unconventional computational mechanism (harmonic resonance). This highlights all the more the need for a general paper to emphasize the shortcomings of the current conventional neural network approach, to allow for the publication of alternative modeling approaches.