Novel View Synthesis via Depth-guided Skip Connections

TL;DR

This paper presents a novel view synthesis method that uses depth-guided skip connections in an encoder-decoder architecture to produce structurally accurate and detail-preserving images from a single source view.

Contribution

It introduces a depth-guided skip connection mechanism to improve detail preservation and reduce distortions in single-image novel view synthesis.

Findings

Reduces distortions compared to flow-based methods.

Preserves texture details effectively.

Achieves structurally consistent novel views.

Abstract

We introduce a principled approach for synthesizing new views of a scene given a single source image. Previous methods for novel view synthesis can be divided into image-based rendering methods (e.g. flow prediction) or pixel generation methods. Flow predictions enable the target view to re-use pixels directly, but can easily lead to distorted results. Directly regressing pixels can produce structurally consistent results but generally suffer from the lack of low-level details. In this paper, we utilize an encoder-decoder architecture to regress pixels of a target view. In order to maintain details, we couple the decoder aligned feature maps with skip connections, where the alignment is guided by predicted depth map of the target view. Our experimental results show that our method does not suffer from distortions and successfully preserves texture details with aligned skip connections.