Due to large variations in shape, appearance, and viewing conditions, objectrecognition is a key precursory challenge in the fields of object manipulationand robotic/AI visual reasoning in general. Recognizing object categories,particular instances of objects and viewpoints/poses of objects are threecritical subproblems robots must solve in order to accurately grasp/manipulateobjects and reason about their environments. Multi-view images of the sameobject lie on intrinsic low-dimensional manifolds in descriptor spaces (e.g.visual/depth descriptor spaces). These object manifolds share the same topologydespite being geometrically different. Each object manifold can be representedas a deformed version of a unified manifold. The object manifolds can thus beparameterized by its homeomorphic mapping/reconstruction from the unifiedmanifold. In this work, we develop a novel framework to jointly solve the threechallenging recognition sub-problems, by explicitly modeling the deformationsof object manifolds and factorizing it in a view-invariant space forrecognition. We perform extensive experiments on several challenging datasetsand achieve state-of-the-art results.
展开▼
