Deep Multi Depth Panoramas for View Synthesis

TL;DR

This paper introduces Multi Depth Panoramas (MDPs), a novel scene representation for view synthesis from 360-degree panoramas, capable of handling large translations and view-dependent effects more effectively than previous methods.

Contribution

The paper presents a deep learning approach to reconstruct MDPs from multi-camera panoramas, enabling high-quality view synthesis with a more compact scene representation.

Findings

MDPs outperform previous RGBD-based methods in view synthesis quality.

The approach effectively handles large translations and view-dependent effects.

Experiments show improved efficiency and realism in synthesized views.

Abstract

We propose a learning-based approach for novel view synthesis for multi-camera 360$^{\circ}$ panorama capture rigs. Previous work constructs RGBD panoramas from such data, allowing for view synthesis with small amounts of translation, but cannot handle the disocclusions and view-dependent effects that are caused by large translations. To address this issue, we present a novel scene representation - Multi Depth Panorama (MDP) - that consists of multiple RGBD$\alpha$ panoramas that represent both scene geometry and appearance. We demonstrate a deep neural network-based method to reconstruct MDPs from multi-camera 360$^{\circ}$ images. MDPs are more compact than previous 3D scene representations and enable high-quality, efficient new view rendering. We demonstrate this via experiments on both synthetic and real data and comparisons with previous state-of-the-art methods spanning both learning-based approaches and classical RGBD-based methods.