3D60 Dataset-360 度相机图像数据集

3D60是在各种360o视觉研究工作的背景下生成的集合数据集[1]，[2]，[3]。它包括来自现实和合成的大型3D数据集（Matterport3D [4]，Stanford2D3D [5]，SunCG [6]）的场景的多模式立体渲染。 现代3D视觉的进步依赖于数据驱动的方法，因此依赖于任务特定的带注释的数据集。特别是对于诸如深度和曲面估计之类的几何推理任务，高质量数据的收集非常具有挑战性，昂贵且费力。尽管对传统的针孔相机已经做出了巨大的努力，但对于全向针孔相机却不能说相同。我们的3D60数据集填补了数据驱动的球形3D视觉中的一个非常重要的空白，更具体地说，是单眼和立体密集深度和表面估计。我们通过利用在提供室内空间的合成和真实扫描3D数据集并通过光线跟踪重新使用它们以产生高质量，带注释的球形全景图方面所做的努力而起源。 我们提供了以下三种不同的方式，以下是相应的数据格式，无效值（由于不完善的扫描，在渲染过程中出现孔洞）在方括号中表示。 我们的球形全景图是使用为Matterport3D和Stanford2D3D提供的摄影机姿势生成的，而对于SunCG，我们是从每个建筑物边界框的中心进行渲染的，这导致了渲染伪像，从而导致了许多无效的渲染。 The 3D60 data have been generated during distinct works and thus, depending on which subset (i.e. modalities and/or placements) are used, please cite the corresponding papers as follows: @inproceedings{zioulis2018omnidepth, title={Omnidepth: Dense depth estimation for indoors spherical panoramas}, author={Zioulis, Nikolaos and Karakottas, Antonis and Zarpalas, Dimitrios and Daras, Petros}, booktitle={Proceedings of the European Conference on Computer Vision (ECCV)}, pages={448--465}, year={2018} } Please direct any questions related to the dataset and tools to [e-mail] or post a GitHub issue.

3D60 is a collective dataset generated in the context of various 360° visual research works [1], [2], [3]. It includes multimodal stereo renderings of scenes sourced from large-scale real and synthetic 3D datasets: Matterport3D [4], Stanford2D3D [5], and SunCG [6].

Advances in modern 3D vision rely on data-driven methods, and thus on task-specific annotated datasets. Particularly for geometric reasoning tasks such as depth and surface estimation, collecting high-quality data is highly challenging, costly, and labor-intensive. While tremendous efforts have been made for traditional pinhole cameras, the same cannot be said for omnidirectional pinhole cameras. Our 3D60 dataset fills a critical gap in data-driven spherical 3D vision, specifically for monocular and stereo dense depth and surface estimation.

Our work originates from leveraging prior efforts in providing synthetic and real-scanned 3D datasets of indoor spaces, and reusing them via ray tracing to produce high-quality, annotated spherical panoramic images.

We provide three distinct modalities, with corresponding data formats specified below. Invalid values (resulting from imperfect scans or holes appearing during rendering) are denoted in square brackets.

Our spherical panoramic images are generated using the camera poses provided for Matterport3D and Stanford2D3D, while for SunCG we render from the center of each building's bounding box, which leads to rendering artifacts and thus numerous invalid renderings.

The 3D60 dataset has been generated in the context of distinct prior works; therefore, depending on the subset (i.e., modalities and/or placements) being used, please cite the corresponding paper as follows:
@inproceedings{zioulis2018omnidepth, title={Omnidepth: Dense depth estimation for indoors spherical panoramas}, author={Zioulis, Nikolaos and Karakottas, Antonis and Zarpalas, Dimitrios and Daras, Petros}, booktitle={Proceedings of the European Conference on Computer Vision (ECCV)}, pages={448--465}, year={2018}
}

Please direct any questions regarding the dataset and tools to [e-mail] or post a GitHub issue.

搜集汇总

背景与挑战

背景概述

3D60 Dataset是一个27.2G大小的360度相机图像数据集，包含来自真实和合成3D数据集的室内场景多模式立体渲染，用于支持球形3D视觉研究，特别是单眼和立体密集深度及表面估计任务。数据集结合了Matterport3D、Stanford2D3D和SunCG等来源的数据，提供了多种数据格式。