MID (Matching In the Dark)

MID数据集是由东北大学GSIS、RIKEN中心AIP和Socionext公司合作创建的，专注于极低光照条件下的图像匹配问题。该数据集包含10,584张图像，涵盖54个室内和54个室外场景，每场景从两个视角拍摄，共49种曝光设置。数据集设计用于评估图像匹配方法在低光照条件下的性能，特别是利用RAW格式图像的高精度信息。创建过程中，通过不同的曝光时间和ISO设置捕捉图像，以评估不同图像增强和匹配方法的效果。MID数据集的应用领域包括结构从运动(SfM)和视觉SLAM，旨在解决极低光照环境下图像匹配的挑战。

The MID dataset was collaboratively developed by the Graduate School of Information Sciences (GSIS) of Tohoku University, the RIKEN Center for Advanced Intelligence Project (AIP), and Socionext Inc., focusing on the problem of image matching under extremely low-light conditions. It contains 10,584 images covering 54 indoor and 54 outdoor scenarios, with two viewpoints captured for each scenario and a total of 49 exposure settings. The dataset is designed to evaluate the performance of image matching methods in low-light environments, particularly by leveraging the high-precision information from RAW-format images. During its creation, images were captured with varying exposure times and ISO settings to assess the effectiveness of different image enhancement and matching approaches. The MID dataset has applications in Structure from Motion (SfM) and visual SLAM, aiming to address the challenges of image matching in extremely low-light environments.

搜集汇总

背景与挑战

背景概述

MID数据集是一个专注于极低光照条件下图像匹配评估的专用数据集，由东北大学GSIS、RIKEN中心AIP和Socionext公司合作创建。它包含10,584张图像，覆盖54个室内和54个室外场景，每场景从两个视角拍摄并采用49种曝光设置，利用RAW格式图像的高精度信息来评估图像增强和匹配方法的性能。该数据集旨在解决极低光照环境下的图像匹配挑战，主要应用于结构从运动(SfM)和视觉SLAM领域。