RGB-MS

通过解决立体匹配对应关系，我们解决了注册具有非常不同分辨率的同步彩色 (RGB) 和多光谱 (MS) 图像的问题。有目的地，我们介绍了一种新颖的rgb-ms数据集，该数据集在室内环境中构建了13个不同的场景，并以视差图的形式提供了总共34个带有半密集，高分辨率地面真相标签的图像对。为了解决该任务，我们提出了一种深度学习体系结构，该体系结构通过利用仅在训练数据采集期间才需要的其他RGB相机来以自我监督的方式进行培训。

By addressing stereo matching correspondence, we solve the problem of registering synchronized color (RGB) and multispectral (MS) images with drastically different resolutions. To this end, we introduce a novel RGB-MS dataset, which consists of 13 distinct indoor scenes and provides a total of 34 image pairs equipped with semi-dense, high-resolution ground truth labels in the form of disparity maps. To tackle this task, we propose a deep learning architecture that is trained in a self-supervised manner by leveraging an additional RGB camera which is only necessary during training data collection.

搜集汇总

背景与挑战

背景概述

RGB-MS是一个用于解决立体匹配对应关系的多光谱图像数据集，包含13个室内场景的34个图像对，提供半密集、高分辨率的地面真相标签。该数据集由博洛尼亚大学于2022年发布，数据量为76.2GB，采用CC BY-NC 4.0许可证。