Regarding the 1995 paper:

Although Canavagh realizes that we should focus on internal representations, he provides no additional insights regarding the matter. If the thesis that we are doing viewpoint-dependent pattern-recognition is correct, then it must explain how we are able to visually recognize
such a large number of objects and understand novel(seen for the first time) objects.

Clearly some hierarchical organization of objects is necessary, and the problem of how a human's experience of the visual world interacts with this internal representation of objects must be addressed. The problem of defining an internal representation of objects that accounts for
reasoning about parts of objects, hierarchical classes of objects, and learn about novel objects given little training data should not be under-estimated. Like mentioned before, we want a representation that can interface with the planning component of the brain; in reality, what we want is deeper than a theory of vision, we want a way to efficienty represent the world internally and reason about it. We should address the issues of viewpoint-dependent recognition and hierarchical representation in tandem.