OASIS: A Large-Scale Dataset for Single Image 3D in the Wild

TL;DR

OASIS is a large-scale dataset with detailed 3D annotations for 140,000 images, designed to advance single-image 3D understanding in real-world scenarios.

Contribution

The paper introduces OASIS, a comprehensive dataset with detailed 3D surface annotations for a vast number of images, enabling improved research in single-image 3D tasks.

Findings

Models trained on OASIS perform better on 3D tasks.

OASIS enables training of models that generalize well to real-world images.

The dataset facilitates significant progress in single-image 3D understanding.

Abstract

Single-view 3D is the task of recovering 3D properties such as depth and surface normals from a single image. We hypothesize that a major obstacle to single-image 3D is data. We address this issue by presenting Open Annotations of Single Image Surfaces (OASIS), a dataset for single-image 3D in the wild consisting of annotations of detailed 3D geometry for 140,000 images. We train and evaluate leading models on a variety of single-image 3D tasks. We expect OASIS to be a useful resource for 3D vision research. Project site: https://pvl.cs.princeton.edu/OASIS.