Matterport3D 360° RGBD

Matterport3D 360 ° RGBD数据集是Matterport3D的扩展，包含训练和验证高分辨率360单目深度估计模型的数据。数据被构造在属于90个不同建筑物的90个文件夹中，总共存储9684个样本。数据集的每个样本由4个文件组成: RGB等矩形360图像 (.png)，其深度地面真相 (.dpt)，深度地面真相 (.png) 的可视化和相机到世界的外部参数图像 (.txt) 保存为7个参数: 3表示相机中心，最后4表示XYWZ旋转四元数。

The Matterport3D 360° RGBD dataset is an extension of the Matterport3D dataset, which contains data for training and validating high-resolution 360° monocular depth estimation models. The dataset is organized into 90 folders corresponding to 90 distinct buildings, storing a total of 9684 samples across all folders. Each sample in the dataset consists of four files: an RGB equirectangular 360° image in .png format, the depth ground truth in .dpt format, a visualization of the depth ground truth in .png format, and a .txt file that saves the camera-to-world extrinsic parameters. The 7 parameters stored in the .txt file are arranged as follows: the first three values represent the camera center, while the last four values denote the XYWZ-formatted rotation quaternion.

搜集汇总

背景与挑战

背景概述

Matterport3D 360° RGBD数据集是Matterport3D的扩展，专门用于训练和验证高分辨率360单目深度估计模型。它包含90个建筑物的9684个样本，每个样本提供RGB图像、深度地面真相和相机参数，数据格式丰富且规模较大（77.0GB）。该数据集由巴斯大学于2022年发布，采用CC BY 4.0许可证，适用于计算机视觉领域的深度估计研究。