TAS-NIR

TAS-NIR数据集由德国联邦国防军大学慕尼黑分校自主系统技术研究所创建，包含209对精细语义分割的可见光和近红外图像，用于非结构化户外环境的地面和植被类型分割。该数据集通过春季、夏季和秋季在不同非结构化户外环境中驾驶的调查车辆上安装的两个相机记录。TAS-NIR数据集主要用于验证和测试，旨在通过融合预训练的卷积神经网络输出和手工制作的可见光+近红外特征，提高非结构化户外环境中自主驾驶的精细语义分割性能。

The TAS-NIR dataset was created by the Institute of Autonomous Systems Technology, Bundeswehr University Munich. It contains 209 pairs of fine semantic-segmented visible light and near-infrared (NIR) images for ground and vegetation type segmentation in unstructured outdoor environments. This dataset was collected using two cameras mounted on a survey vehicle driving across various unstructured outdoor environments during spring, summer, and autumn. The TAS-NIR dataset is primarily intended for validation and testing, with the goal of improving the performance of fine semantic segmentation for autonomous driving in unstructured outdoor environments by fusing the outputs of pre-trained convolutional neural networks (CNNs) and hand-crafted visible light and NIR features.

搜集汇总

背景与挑战

背景概述

TAS-NIR是一个用于非结构化户外环境中精细语义分割的数据集，包含209个可见光和近红外（VIS-NIR）图像对，并带有地形和植被的详细标注。该数据集由慕尼黑联邦国防军大学自主系统技术研究所发布，适用于非商业用途的研究，采用署名-非商业性-相同方式共享3.0许可证。