User Manual Privacy Policy Disclaimer Contact us
  Advanced SearchBrowse





IGNOR: Image-guided Neural Object Rendering


Theobalt,  Christian
Computer Graphics, MPI for Informatics, Max Planck Society;


(Supplementary material)

Fulltext (public)

(Preprint), 5MB

Supplementary Material (public)
There is no public supplementary material available

Thies, J., Zollhöfer, M., Theobalt, C., Stamminger, M., & Nießner, M. (2018). IGNOR: Image-guided Neural Object Rendering. Retrieved from http://arxiv.org/abs/1811.10720.

Cite as: http://hdl.handle.net/21.11116/0000-0002-F7EB-F
We propose a new learning-based novel view synthesis approach for scanned objects that is trained based on a set of multi-view images. Instead of using texture mapping or hand-designed image-based rendering, we directly train a deep neural network to synthesize a view-dependent image of an object. First, we employ a coverage-based nearest neighbour look-up to retrieve a set of reference frames that are explicitly warped to a given target view using cross-projection. Our network then learns to best composite the warped images. This enables us to generate photo-realistic results, while not having to allocate capacity on `remembering' object appearance. Instead, the multi-view images can be reused. While this works well for diffuse objects, cross-projection does not generalize to view-dependent effects. Therefore, we propose a decomposition network that extracts view-dependent effects and that is trained in a self-supervised manner. After decomposition, the diffuse shading is cross-projected, while the view-dependent layer of the target view is regressed. We show the effectiveness of our approach both qualitatively and quantitatively on real as well as synthetic data.