Tmsk building dataset

The obtained raw data were stitched and radiometrically corrected. The buildings were labeled by manual decoding, and the building area of less than 5% was removed. The final Tmsk Building dataset is 1,323, randomly divided into 1,058 training sets, 132 validation sets, and 133 test sets. To avoid overfitting the network, the data are enhanced by random inversion, image noise addition, and increasing image brightness.

所获取的原始数据已完成拼接与辐射校正。建筑物通过人工解译完成标注，并移除了面积占比低于5%的建筑区域。最终的Tmsk Building数据集共计1323份样本，经随机划分为训练集1058份、验证集132份与测试集133份。为避免网络过拟合，我们通过随机翻转、添加图像噪声以及提升图像亮度的方式对数据进行增强。

搜集汇总

背景与挑战

背景概述

Tmsk building dataset是一个用于建筑物提取的数据集，包含1,323个经过拼接和辐射校正的高分辨率可见光图像，手动标注建筑物并去除面积小于5%的建筑区域。数据集被随机划分为1,058个训练集、132个验证集和133个测试集，并通过随机反转、添加图像噪声和增加图像亮度等方法进行数据增强以避免网络过拟合。