NAVER LABS Localization Datasets

NAVER LABS 定位数据集是 5 个新的室内数据集，用于在具有挑战性的现实环境中进行视觉定位。他们是在韩国首尔的一个大型购物中心和一个大型地铁站使用由 10 台摄像机和 2 台激光扫描仪组成的专用测绘平台拍摄的。为了获得准确的地面实况相机位姿，我们使用了一个强大的 LiDAR SLAM，它提供了初始位姿，然后使用一种基于运动结构的新型优化来改进这些位姿。数据集以 kapture 格式提供，包含大约 130k 图像以及用于训练和验证的 6DoF 相机姿势。我们还为训练图像提供基于稀疏激光雷达的深度图。测试集的位姿被保留以不偏向基准。

The NAVER LABS Localization Dataset consists of 5 new indoor datasets designed for visual localization in challenging real-world environments. These datasets were collected at a large shopping mall and a major subway station in Seoul, South Korea, using a dedicated mapping platform equipped with 10 cameras and 2 laser scanners. To obtain accurate ground-truth camera poses, we employed a robust LiDAR SLAM system to generate initial poses, followed by a novel structure-from-motion-based optimization to refine these poses. The datasets are provided in the kapture format, containing approximately 130,000 images and 6DoF camera poses for training and validation. We also provide sparse LiDAR-based depth maps for the training images. The poses of the test set are withheld to avoid biasing the benchmark.

搜集汇总

背景与挑战

背景概述

NAVER LABS Localization Datasets是一个用于视觉定位的室内数据集集合，包含5个在韩国首尔购物中心和地铁站采集的数据集，使用多摄像头和激光扫描仪平台。该数据集提供约13万张图像、6DoF相机位姿及稀疏激光雷达深度图，适用于图像检索和基于图像的定位任务，测试集位姿保留以支持基准评估。