FudanCVL/ObjectPose9D

ObjectPose9D数据集是一个用于对象姿态估计的数据集，包含来自不同来源的数据样本。每个样本包括四个属性：来源数据集名称（source）、文本描述或提示（prompt）、图像数据（image，以二进制存储）和CNOCS Map（map，以EXR格式的二进制存储）。对于Cityscapes子集，由于许可限制，image字段仅包含相对路径，用户需要自行下载原始Cityscapes数据集。对于其他来源，image字段包含实际的二进制图像数据。数据集还提供了加载和可视化CNOCS Maps的代码示例。

The ObjectPose9D dataset is designed for object pose estimation, containing data samples from various sources. Each sample includes four attributes: source dataset name (source), text description or prompt (prompt), image data (image, stored as binary), and CNOCS Map (map, stored as binary in EXR format). For the Cityscapes subset, due to license restrictions, the image field contains only the relative path within the original dataset, requiring users to download the original Cityscapes dataset separately. For other sources, the image field contains the actual binary image data. The dataset also provides code examples for loading the dataset and visualizing CNOCS Maps.