Do viewpoint-dependent mechanisms generalize across members of a class?

Area of Science:

• Cognitive Science
• Computer Vision
• Psychology

Background:

• Object recognition theories often assume abstract representations.
• Previous research focused on exemplar-specific tasks for viewpoint-dependent representations.
• The ability to recognize novel instances of familiar classes is a key aspect of human vision.

Purpose of the Study:

• To investigate if viewpoint-specific representations generalize across members of an object class.
• To determine if image-based representations support generalization in categorical tasks.
• To explore the role of familiarity and geometric distinctiveness in cross-instance recognition.

Main Methods:

• Three experiments tested generalization of familiarity with novel 3D objects.
• Participants learned specific members of object classes.
• Generalization to new, visually similar objects was assessed under varying viewpoint conditions.

Main Results:

• Familiarity with some objects in a class generalized to other objects in that class in a viewpoint-dependent manner.
• Generalization was influenced by the degree of familiarity and viewpoint-specific geometric distinctiveness.
• This generalization effect was limited to visually similar objects within the learned context.

Conclusions:

• Image-based object representations are viewpoint-dependent.
• These viewpoint-specific representations can generalize across members of perceptually-defined object classes.
• Object classes may be represented as clusters of visually similar, viewpoint-specific representations, supporting a novel approach to recognition.