Functionality can be defined as the “purpose for which an object is designed or used”. Thus, an understanding of functionality can reveal fundamental characteristics of an object, such as actions that the object can support and its possible uses, for example, “a table allows objects to rest on its surface”, or “a drawer can open/close to provide storage”. Such insights on the functionality of an object can be useful for various computer graphics and vision applications, such as classification, retrieval, modeling, and animation. Moreover, although the analysis of the shape of an object can convey a significant amount of information about the object’s intended functionality, a more complete discovery of functionality will necessarily involve the analysis of interactions of the object with other objects, and possibly also the interaction that happens among parts of the object.

In this talk, I will discuss some of our recent work where we analyze the functionality of objects based on interactions, in order to learn models that represent different functionalities. After outlining and motivating the problem of functionality analysis, I will discuss a first line of work where we build a model of static functionality for an object category. The functionality model is learned from objects provided in example scenes that demonstrate the static interactions that appear in the category. The model can then predict whether an unknown object possesses the functionality of the category. Moreover, I will discuss a second approach where we analyze the mobility of object parts, which is relevant to the analysis of dynamic functionality that involves the motion and interaction between parts of an object. The part mobility model is learned from example motions and can be used to predict the mobility of a given pair of parts. I will finalize the talk by discussing challenges and future directions in the analysis of functionality based on object interactions.