FlOoD: Synthetic Dataset for Semantic Segmentation of Waterbodies in Out-of-Distribution situations

The dataset is split into two directories: train, and test. The folder structure of each of the (train, test) directories is as follows: A directory corresponds to a 3D scene. For each scene, there are directories that correspond to a camera placed in the 3D scene. For each camera, there are directories that include the RGB frames (images) and corresponding directories (masks) that contain the respective binary segmentation masks. A metadata.json file contains the attribute values for each frame.

本数据集划分为训练集（train）与测试集（test）两个目录。训练集与测试集的文件夹结构一致，均遵循如下规则：每个子目录对应一个三维场景（3D scene）。针对每个三维场景，其下设有若干子目录，分别对应布置于该三维场景中的一台相机。针对每台相机，其下设有两类子目录：一类用于存储RGB帧图像（RGB frames），另一类为掩码目录（masks），用于存放对应的二值分割掩码（binary segmentation masks）。目录中包含metadata.json文件，该文件存储了每一帧的属性参数值。