Novel View Synthesis from Single Images via Point Cloud Transformation

TL;DR

This paper introduces a method for novel view synthesis from a single image by estimating a point cloud to represent object geometry, then using image completion to generate dense views, validated on the ShapeNet benchmark.

Contribution

The approach combines point cloud estimation from a single image with end-to-end training using warping techniques, enabling explicit 3D shape use for view synthesis without depth supervision.

Findings

Effective on ShapeNet benchmark

Produces dense novel views from single images

End-to-end training without depth supervision

Abstract

In this paper the argument is made that for true novel view synthesis of objects, where the object can be synthesized from any viewpoint, an explicit 3D shape representation isdesired. Our method estimates point clouds to capture the geometry of the object, which can be freely rotated into the desired view and then projected into a new image. This image, however, is sparse by nature and hence this coarse view is used as the input of an image completion network to obtain the dense target view. The point cloud is obtained using the predicted pixel-wise depth map, estimated from a single RGB input image,combined with the camera intrinsics. By using forward warping and backward warpingbetween the input view and the target view, the network can be trained end-to-end without supervision on depth. The benefit of using point clouds as an explicit 3D shape for novel view synthesis is experimentally validated on the 3D ShapeNet benchmark. Source code and data will be available at https://lhoangan.github.io/pc4novis/.