Novel View Synthesis with Pixel-Space Diffusion Models

TL;DR

This paper introduces a pixel-space diffusion model for novel view synthesis from a single image, achieving superior results and better generalization by leveraging single-view datasets and innovative encoding strategies.

Contribution

It adapts diffusion models for end-to-end NVS, introduces a new training scheme with single-view data, and evaluates geometric encoding methods, showing their limited impact compared to model improvements.

Findings

Outperforms previous SOTA methods in NVS tasks.

Single-view training improves generalization to out-of-domain scenes.

Geometric encoding methods have minor effects compared to model enhancements.

Abstract

Synthesizing a novel view from a single input image is a challenging task. Traditionally, this task was approached by estimating scene depth, warping, and inpainting, with machine learning models enabling parts of the pipeline. More recently, generative models are being increasingly employed in novel view synthesis (NVS), often encompassing the entire end-to-end system. In this work, we adapt a modern diffusion model architecture for end-to-end NVS in the pixel space, substantially outperforming previous state-of-the-art (SOTA) techniques. We explore different ways to encode geometric information into the network. Our experiments show that while these methods may enhance performance, their impact is minor compared to utilizing improved generative models. Moreover, we introduce a novel NVS training scheme that utilizes single-view datasets, capitalizing on their relative abundance compared to their multi-view counterparts. This leads to improved generalization capabilities to scenes with out-of-domain content.