A Real World Dataset for Multi-view 3D Reconstruction

TL;DR

This paper introduces a comprehensive real-world dataset with 998 3D models and 847,000 images, annotated for camera and object poses, to advance multi-view 3D reconstruction research.

Contribution

It provides a new large-scale, real-world dataset with detailed annotations specifically designed for multi-view 3D reconstruction tasks.

Findings

Dataset enables improved learning for 3D shape reconstruction

Annotations facilitate accurate pose estimation

Baseline evaluations demonstrate dataset's utility

Abstract

We present a dataset of 998 3D models of everyday tabletop objects along with their 847,000 real world RGB and depth images. Accurate annotations of camera poses and object poses for each image are performed in a semi-automated fashion to facilitate the use of the dataset for myriad 3D applications like shape reconstruction, object pose estimation, shape retrieval etc. We primarily focus on learned multi-view 3D reconstruction due to the lack of appropriate real world benchmark for the task and demonstrate that our dataset can fill that gap. The entire annotated dataset along with the source code for the annotation tools and evaluation baselines is available at http://www.ocrtoc.org/3d-reconstruction.html.