Abstract
The existence of cue-invariant neural mechanisms for representing visual depth and shape has been hypothesized, though the data favoring this hypothesis is currently sparse. We have found that performance improvements by human observers on a visual slant discrimination task transferred from training conditions in which planar surfaces were defined solely by a stereo cue to testing conditions in which surfaces were defined solely by texture and motion cues, and vice versa. Methods: On each training trial, a subject viewed two consecutive presentations of planar surfaces slanted in depth, and judged whether the surface slants were the same or different. An auditory feedback signal indicated the correctness of a subject's judgment. For one group of subjects, surfaces were depicted via a stereo cue only—small dots were placed on a surface and rendered from left-eye and right-eye viewpoints. For a second group of subjects, surfaces were depicted via texture and motion cues only—texture elements were placed on a surface, and the surface rotated back and forth around a vertical axis. Each group performed training trials during three experimental sessions. Before and after training, subjects performed test trials during which they performed the same/different judgment task both when surfaces were defined by a stereo cue and when they were defined by texture and motion cues. Results: A comparison of performances on post-test versus pre-test trials shows that subjects trained in one cue condition showed significantly improved performance when tested in the other cue condition. In addition, performance improvements were not limited to the specific slant used during training, but were also evident when surfaces were centered at a novel slant. Conclusions: Training with one cue led to performance improvements with another cue, suggesting that this improvement is due, at least in part, to adaptation of neural mechanisms that represent surface slant in a cue-invariant manner.