The point is that the distinction between reality and illusion presupposes folk psychological categories (mental representations that can be right or wrong), to which the eliminativist isn't entitled.

Notice what your phrase "only in our perception" assumes. To say that the rod is bent only in our perception and not in its "actual" physical manifestation is to appeal to the illusion-reality distinction. But that's a distinction that's supposed to be eliminated in favour of a strictly scientific account of causes and effects.

If we restrict ourselves to talk of causality, all we can say is that the light produces visual data in a brain, of a wavy thing in the water. Even to add that we can figure out what's really going on, by drawing inferences that the rod is really straight and that light and the water aren't powerful enough to bend that object is again to presuppose illusion-reality and mental representations that can be right or wrong, so we'd be back at folk psychology.

Bakker wants to get around this by distinguishing instead between high-informational and low-informational knowledge, but that's a fig leaf. Information theory assumes that signals are sent and received by minds with intentions that are interested in the meaning of signals rather than in noise.

Otherwise, the information contents in the appearance of the wavy rod in water or of the straight rod out of water are equally high; they're just indications of different systems and background conditions. The difference is that we may be more interested in the straight rod (the "real" object) than in its wavy appearance in water, and that would again amount to an appeal to the folk conception of minds.