2D-3D Match Dataset 数据集

2D-3D Match Dataset 是一个通过利用来自 RGB-D 扫描的几个 3D 数据集实现 2D-3D 对应的数据集。数据集使用来自 SceneNN 和 3DMatch 的数据。训练数据集由 110 个 RGB-D 扫描组成，其中 56 个场景来自 SceneNN，54 个场景来自 3DMatch，给定从 3D 点云随机采样的 3D 点，从不同扫描视图中提取一组 3D 补丁。该数据集采用一种学习局部跨域描述符的新方法，以用于二维图像和三维点云匹配。数据集包含从公开的 RGB-D 场景中获得 140 万个具有不同照明条件和设置的 2D-3D 对应关系

The 2D-3D Match Dataset is a specialized dataset for establishing 2D-3D correspondences, which leverages multiple 3D datasets derived from RGB-D scans. It employs data sourced from SceneNN and 3DMatch. The training dataset consists of 110 RGB-D scans, including 56 scenes from SceneNN and 54 scenes from 3DMatch. Given 3D points randomly sampled from 3D point clouds, a set of 3D patches is extracted from different scan viewpoints. This dataset adopts a novel method for learning local cross-domain descriptors tailored for 2D image and 3D point cloud matching. The dataset contains 1.4 million 2D-3D correspondences acquired from public RGB-D scenes with varying lighting conditions and acquisition settings.

搜集汇总

背景与挑战

背景概述

2D-3D Match Dataset 是一个用于二维图像和三维点云匹配的数据集，基于SceneNN和3DMatch的RGB-D扫描构建，包含140万个具有不同照明条件和设置的2D-3D对应关系。该数据集大小为30.95 GB，旨在支持学习局部跨域描述符，适用于图像识别任务。