Abstract
In three experiments, Bennett (VSS, 2003) found a stereo advantage in generalizing to unfamiliar views in a sequential matching task in which viewers matched the identity of shaded tube-like objects. We extended these results in an experiment in which subjects identified rotating wire-frame objects at the individual level under stereo and nonstereo viewing conditions; the new experiment tapped longer term memory representations, and provided subjects with multiple views of the training and test objects. During training these objects were presented from a particular view. After training, we tested subjects' ability to recognize the learned targets presented from familiar and unfamiliar views, where training and testing were conducted under either stereo or nonstereo viewing conditions. Though performance was view dependent under both stereo and nonstereo viewing (as in Bennett, VSS, 2003), subjects generalized better to unfamiliar orientations under stereo viewing. Taken together with Bennett (VSS, 2003), these results strongly argue against strictly 2D image-based models of object recognition, at least for the stimuli and recognition tasks used, and they suggest that observers used representations that contained view-specific local depth information.