Mila Simulated Floods

Mila Simulated Floods Dataset 是一个 1.5 平方公里的虚拟世界，使用 Unity3D 游戏引擎，包括城市、郊区和农村地区。
城市环境包含摩天大楼、大型建筑物和道路，以及交通物品和车辆等物体。乡村环境由草山、森林和山脉组成，房屋和教堂等其他建筑物稀疏，没有道路。农村和城市地区占我们虚拟世界的 1 平方公里。
郊区环境是一个0.5平方公里的住宅区，有许多带有前院的独立房屋。
为了收集模拟数据集，我们使用以下方式从 2000 个视点捕获了洪水对之前和之后的数据：
非泛洪 RGB 图像、深度图、分割图
淹没的 RGB 图像，淹没区域的二进制掩码，分割图
相机放置在离地面约 1.5 度的位置，视野为 120 度，图像分辨率为 1200 x 900。在每个视点，我们拍摄了 10 张照片，通过稍微改变相机的位置以扩充数据集。

The Mila Simulated Floods Dataset is a 1.5-square-kilometer virtual world developed using the Unity3D game engine, encompassing urban, suburban, and rural environments. The urban environment features skyscrapers, large-scale buildings, roads, as well as traffic-related objects and vehicles. The rural environment comprises grassy hills, forests, and mountains, with sparse structures such as houses and churches, and no road networks. The combined urban and rural areas cover 1 square kilometer of this virtual world. The suburban environment is a 0.5-square-kilometer residential district with numerous detached houses equipped with front yards. To compile this simulated dataset, we captured pre- and post-flood data from 2000 viewpoints via the following modalities: - Non-flooded RGB images, depth maps, and segmentation masks - Flooded RGB images, binary masks of submerged regions, and segmentation masks. Cameras were positioned approximately 1.5 meters above the ground (corrected for the likely typo of 1.5 degrees to meters, as degrees is an angular unit unsuitable for describing height), with a 120-degree field of view, and the image resolution was set to 1200 × 900. At each viewpoint, we captured 10 photos by slightly adjusting the camera's position to perform data augmentation on the dataset.

搜集汇总

背景与挑战

背景概述

Mila Simulated Floods是一个基于Unity3D构建的1.5平方公里虚拟环境数据集，涵盖城市、郊区和农村区域。它从2000个视点采集了非泛洪和淹没状态的RGB图像、深度图及分割图，图像分辨率为1200 x 900，适用于深度估计、场景分割等计算机视觉任务。