"Indian Cities Change Detection (ICCD) Dataset"

"Indian Cities Change Detection (ICCD) Dataset is a curated high-resolution remote sensing dataset developed for semi-supervised building change detection. It consists of bi-temporal satellite image pairs (im1 and im2) and corresponding change masks (label) for ten Indian cities: Agra, Ambala, Amritsar, Jammu, Kanpur, Kannur, Kargil, Pune, Roorkee, and Sagar. The dataset captures diverse urban, semi-urban, and rural landscapes with temporal variations, including new building construction, demolitions, redevelopment, deforestation, water body changes, and seasonal vegetation transitions. It encompasses newly constructed buildings, demolished structures, and redeveloped sites where existing buildings have been replaced by new establishments.The dataset is organized city-wise, enabling evaluation and generalization across different urban domains. For each city, labeled and unlabeled data have been collected from 2020 to 2024, depending on availability. The labeled set contains im1, im2, and label folders, while the unlabeled set contains only im1 and im2. All images are in RGB format at 1024\u00d71024 resolution. Building changes are annotated in white (255, 255, 255), with unchanged areas in black (0, 0, 0).The ICCD dataset supports applications in:\u2022 Building change detection\u2022 Domain adaptation across diverse geographies\u2022 Semi-supervised learning with city-wise domain shiftsThis dataset provides a comprehensive, real-world benchmark to advance research in remote sensing, change detection, and domain adaptation."

印度城市变化检测（Indian Cities Change Detection, ICCD）数据集是一套专为半监督建筑变化检测任务打造的精选高分辨率遥感数据集。它包含10座印度城市的双时相卫星图像对（im1与im2）以及对应的变化掩码标签（label），涉及城市分别为阿格拉、安巴拉、阿姆利则、查谟、坎普尔、坎努尔、卡吉尔、浦那、鲁尔基与萨加尔。
该数据集覆盖了多样化的城市、半城市与乡村景观，包含各类时序变化场景，涵盖新建建筑施工、建筑拆除、重建、森林砍伐、水体变化以及季节性植被演替等类型。其收录范围包括新建建筑、已拆除建筑以及原有建筑被新建筑替代的重建场地。
该数据集按城市维度进行组织，支持在不同城市域间开展模型评估与泛化能力测试。针对每座城市，我们根据数据可获取性，于2020年至2024年间收集了带标注与无标注数据。带标注数据集包含im1、im2与label三个文件夹，而无标注数据集仅包含im1与im2文件夹。所有图像均为RGB格式，分辨率为1024×1024像素。建筑变化区域以白色（255, 255, 255）标注，未发生变化的区域则以黑色（0, 0, 0）标注。
ICCD数据集可应用于以下场景：
• 建筑变化检测
• 跨多样地理区域的域自适应
• 考虑城市域偏移的半监督学习
本数据集为遥感、变化检测以及域自适应领域的研究提供了一套全面且贴合真实场景的基准测试集。