SZTAKI AirChange Benchmark set

这个基准集包含13个航拍图像对，尺寸为952x640，分辨率为1.5m/像素和二进制变化掩模 (手工绘制)，用于评估出版物 [1] 和 [2]。每个记录包含一对初步注册的输入图像和一个 “相关” 变化的掩模。输入图像分别为5,7。23年的时间差。在生成变更面具期间，我们将以下差异视为相关变更 :( a) 新建区域 (b) 建筑运营 (c) 种植大树群 (d) 新鲜耕地 (e) 建筑前的基础。请注意，地面真相不包含更改分类，仅包含二进制更改-每个像素没有更改决策。

This benchmark dataset contains 13 pairs of aerial images with a size of 952×640, a resolution of 1.5 meters per pixel, and hand-drawn binary change masks, which are used for evaluating publications [1] and [2]. Each record consists of a pair of co-registered input images and a mask of "relevant" changes. The two input images in each pair have a temporal difference of either 5 years or 23 years. During the generation of the change masks, the following differences are regarded as relevant changes: (a) newly built areas; (b) building construction operations; (c) planting clusters of large trees; (d) newly cultivated farmland; (e) foundations laid prior to building construction. Please note that the ground truth does not include change classification, only binary change labels, where each pixel is assigned a binary decision of either changed or unchanged.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于变化检测评估的基准集，包含13对航拍图像和二进制变化掩模，图像分辨率为1.5米/像素，时间跨度5到23年，专注于新建区域、建筑运营等特定变化类型。它由匈牙利科学院于2008年发布，主要用于遥感图像分析研究，但掩模仅提供像素级二进制变化信息，无详细分类。