Virtual View Networks for Object Reconstruction

TL;DR

This paper introduces a method to synthesize virtual views for object reconstruction using networks of similar objects, enabling accurate shape recovery from minimal images and extending structure-from-motion applications.

Contribution

It proposes a novel approach to generate virtual views via geodesic computations on object networks, improving reconstruction from limited images.

Findings

Accurate shape reconstruction from a single image on PASCAL VOC data

Method extends the applicability of structure-from-motion techniques

Robustness improvements in factorization-based reconstruction

Abstract

All that structure from motion algorithms "see" are sets of 2D points. We show that these impoverished views of the world can be faked for the purpose of reconstructing objects in challenging settings, such as from a single image, or from a few ones far apart, by recognizing the object and getting help from a collection of images of other objects from the same class. We synthesize virtual views by computing geodesics on novel networks connecting objects with similar viewpoints, and introduce techniques to increase the specificity and robustness of factorization-based object reconstruction in this setting. We report accurate object shape reconstruction from a single image on challenging PASCAL VOC data, which suggests that the current domain of applications of rigid structure-from-motion techniques may be significantly extended.