OmniObject3D-beta

OmniObject3D 是一个大规模、多种类、高精度、真实物体扫描的 3D 数据集，能够支持神经渲染、表面重建、3D 生成、点云识别等多种任务的泛化性和鲁棒性的研究，并有利于搭建 2D-3D 理解和算法融合的桥梁。

√ 6,000 个高精 3D 模型
√ 每个物体包括 5 条环绕视频（不同轨迹、灯光条件）
√ > 190 个物体类别
√ 网格 + 纹理、点云、实拍视频、渲染多视角图片、相机位姿
√ 支持多种下游任务

OmniObject3D is a large-scale, multi-category, high-precision 3D dataset composed of real-world object scans. It enables research on the generalization and robustness of diverse tasks including neural rendering, surface reconstruction, 3D generation and point cloud recognition, and facilitates the establishment of a bridge between 2D-3D understanding and algorithm fusion.

√ 6,000 high-precision 3D models
√ Each object includes 5 surround-view videos with different trajectories and lighting conditions
√ Over 190 object categories
√ Provides mesh + texture, point clouds, real-captured videos, rendered multi-view images and camera poses
√ Supports multiple downstream tasks

搜集汇总

背景与挑战

背景概述

OmniObject3D-beta是一个大规模、高精度的真实物体3D扫描数据集，包含6,000个3D模型和超过190个物体类别，每个物体提供5条不同轨迹和灯光条件的环绕视频。该数据集支持神经渲染、表面重建、3D生成和点云识别等多种任务，旨在推动2D-3D理解和算法融合的研究。