Long-term visual localization

视觉定位是估计 6 自由度 (DoF) 相机位姿的问题，从该位姿中获取给定图像相对于参考场景表示。视觉定位是增强、混合和虚拟现实等应用以及机器人技术（例如自动驾驶汽车）的关键技术。为了评估更长时间内的视觉定位，我们提供了基准数据集，旨在评估由季节性（夏季、冬季、春季等）和照明（黎明、白天、日落，夜晚）条件。每个数据集由一组参考图像及其对应的地面实况姿势和一组查询图像组成。为每个数据集提供了一个三角 3D 模型，并且可以用于基于结构的定位方法。为确保结果的公平性和可比性，查询图像的参考姿势被保留，我们提供评估服务来测量姿势准确性。

Visual localization refers to the problem of estimating the 6 Degrees of Freedom (DoF) camera pose from which a given image was captured relative to a reference scene representation. Visual localization is a critical technology for applications such as augmented, mixed, and virtual reality, as well as robotics (e.g., autonomous vehicles). To evaluate visual localization over extended periods, we present a benchmark dataset intended to assess performance under varying seasonal (e.g., summer, winter, spring) and lighting (e.g., dawn, daytime, sunset, night) conditions. Each dataset comprises a set of reference images paired with their corresponding ground-truth poses, as well as a set of query images. A triangular 3D model is provided for each dataset, which can be utilized by structure-based localization methods. To ensure fairness and comparability of experimental results, the ground-truth poses of the query images are withheld, and we offer an evaluation service to measure pose accuracy.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于评估长期视觉定位性能的基准数据集，专注于测试不同季节和光照条件下的6自由度相机位姿估计。它包含参考图像、查询图像、地面实况姿势以及三角3D模型，并提供评估服务以确保结果的公平性和可比性。