Abstract
Models of face coding often assume "opponent" representations in which opposite faces are represented by opposite responses relative to a norm. The norm could be implicit (e.g. equal activity in channels tuned to opposing directions) or explicit (e.g. a null in a single mechanism that responds oppositely to opposing directions, similar to color opponency). Most studies of face aftereffects fail to distinguish between these alternatives because they adapt to a single face. We examined adaptation to sequences of faces that randomly varied over time along one direction but with a fixed mean centered on the norm. Faces were distorted by local expansion or contraction along the vertical (90-270 deg) or horizontal (0-180 deg) axes. Observers were adapted to one axis in the left visual field and the orthogonal axis in the right. They then adjusted the angle of a test face using a staircase until the vertical and horizontal distortions appeared equal (i.e. angles of 45 or 225 deg). Adaptation to each orientation biased the test faces away from the adapting axis – analogous to a tilt-aftereffect within the space. The two test poles rotated in the same way (e.g. clockwise), arguing against a mean bias. Distortion aftereffects for a single face tend to be uniform across the space, suggesting that the tilts are also unlikely to reflect locally stronger biases from more nearby adaptors. Instead, the aftereffects could reflect a loss in contrast sensitivity, and thus adaptation at an explicit opponent site. This contrast adaptation provides an alternative explanation for contingent face-aftereffects, in which opposing aftereffects occur in faces paired with opposing features. These are typically interpreted in terms of separate norms for different face categories, but could also reflect tilt-aftereffects relative to a single norm.
Meeting abstract presented at VSS 2016