I read it as saying they're conventional in the sense that the lines between categories of sensory experience are drawn by consensus, lacking direct access to the experiences of others.

We of course lack direct access to the atomic-scale world as well, but I imagine that's the point -- atomism was a lot more abstract to the likes of Democritus than atomic theory is to us. The underlying physical reality is in a certain sense abstracted away from us, and we reflect that by talking about physical experiences in a conventional way, but those experiences are still rooted in the reality of atoms and space -- or distributions of probability density, if you prefer.