MegaScenes: Scene-Level View Synthesis at Scale

TL;DR

This paper introduces MegaScenes, a large-scale scene-level dataset from Internet photos, and proposes methods to improve view synthesis consistency, enabling better in-the-wild scene generation.

Contribution

The paper creates MegaScenes, a diverse dataset from Internet photos, and enhances scene-level view synthesis methods for more consistent in-the-wild scene generation.

Findings

MegaScenes contains over 100K SfM reconstructions from Internet photos.

Significant improvements in generation consistency over state-of-the-art methods.

Validated effectiveness on diverse real-world scenes.

Abstract

Scene-level novel view synthesis (NVS) is fundamental to many vision and graphics applications. Recently, pose-conditioned diffusion models have led to significant progress by extracting 3D information from 2D foundation models, but these methods are limited by the lack of scene-level training data. Common dataset choices either consist of isolated objects (Objaverse), or of object-centric scenes with limited pose distributions (DTU, CO3D). In this paper, we create a large-scale scene-level dataset from Internet photo collections, called MegaScenes, which contains over 100K structure from motion (SfM) reconstructions from around the world. Internet photos represent a scalable data source but come with challenges such as lighting and transient objects. We address these issues to further create a subset suitable for the task of NVS. Additionally, we analyze failure cases of state-of-the-art NVS methods and significantly improve generation consistency. Through extensive experiments, we validate the effectiveness of both our dataset and method on generating in-the-wild scenes. For details on the dataset and code, see our project page at https://megascenes.github.io.