RIO-D3D

RIO-D3D是一个大规模的室内数据集，由慕尼黑工业大学创建，专注于单目深度预测任务。该数据集包含34000个高质量的渲染深度图，这些深度图来源于RGB-D重建，提供了完整的3D几何信息和高质量的深度数据。数据集的创建过程中，利用了3RScan数据集的1482个RGB-D扫描序列，通过离线处理获取了捆绑调整的相机姿态和纹理映射的3D重建。RIO-D3D数据集不仅适用于深度预测的训练和评估，还支持3D场景完成的任务，旨在解决深度预测中的结构准确性问题，推动深度学习在3D视觉任务中的应用。

RIO-D3D is a large-scale indoor dataset created by the Technical University of Munich, focusing on monocular depth prediction tasks. It contains 34,000 high-quality rendered depth maps derived from RGB-D reconstructions, which provide complete 3D geometric information and high-quality depth data. During its development, 1482 RGB-D scanning sequences from the 3RScan dataset were utilized, with bundle-adjusted camera poses and texture-mapped 3D reconstructions obtained via offline processing. The RIO-D3D dataset supports not only the training and evaluation of depth prediction models but also 3D scene completion tasks, aiming to address the structural accuracy issues in depth prediction and promote the application of deep learning in 3D vision tasks.