Single and sparse view 3D reconstruction by learning shape priors

摘要

In this paper, we aim to reconstruct free-form 3D models from only one or few silhouettes by learning the prior knowledge of a specific class of objects. Instead of heuristically proposing specific regularities and defining parametric models as previous research, our shape prior is learned directly from existing 3D models under a framework based on the Gaussian Process Latent Variable Model (GPLVM). The major contributions of the paper include: (1) a framework for learning the shape prior of the 3D objects, which requires no heuristic of the object, and can be easily generalized to handle various categories of 3D objects and (2) novel probabilistic inference schemes for automatically reconstructing 3D shapes from the silhouette(s) in the single view or sparse views. Qualitative and quantitative experimental results on both synthetic and real data demonstrate the efficacy of our new approach.