IBD（InstanceBuilding 2D）

名为InstanceBuilding的数据集同时包含3D城市场景和无人机图像的建筑物实例注释，这使其具有独特性。我们为608无人机图像注释了16,000屋顶，所有这些都通过相应的高度图进行了增强。我们还为3D城市场景注释了892屋顶/建筑物，其中600多个建筑物与其他建筑物相连。在这样拥挤的城市场景中，实例分割比各种应用中的语义分割具有更显著的 fi can优势。与现有数据集的比较可以在我们的IEEE TGRS论文中找到。用屋顶实例分段掩码注释的高分辨率无人机无人机图像。

The InstanceBuilding dataset is unique in that it contains building instance annotations for both 3D urban scenes and drone-captured images. We annotated 16,000 roofs across 608 drone images, all of which are augmented with corresponding height maps. We also annotated 892 roofs/buildings for 3D urban scenes, with over 600 of these buildings connected to other structures. In such crowded urban scenarios, instance segmentation exhibits more significant advantages over semantic segmentation across various applications. Comparisons with existing datasets can be found in our IEEE TGRS paper. The dataset includes high-resolution drone images annotated with roof instance segmentation masks.

搜集汇总

背景与挑战

背景概述

IBD（InstanceBuilding 2D）是一个包含3D城市场景和无人机图像建筑物实例注释的数据集，规模为4.9GB，包含22k数据项。该数据集为608张无人机图像和892个3D城市场景提供了屋顶/建筑物的实例分割注释，适用于建筑物实例分割任务。