DataDoP
收藏资源简介:
DataDoP数据集是一个大规模的多模态数据集,包含29K个现实世界中的自由移动相机轨迹,深度图和详细的运动标签。数据集由浙江大学、上海人工智能实验室等机构创建,每个样本包括一个镜头级别的相机轨迹,对应的RGBD图像,以及两种类型的轨迹描述:运动描述和导演描述。运动描述仅描述相机运动,而导演描述则包括相机运动、与场景的互动和导演意图。该数据集聚焦于艺术性的自由移动轨迹,能够捕捉导演的创意视野,具有 cinematic 和艺术价值。
The DataDoP Dataset is a large-scale multimodal dataset containing 29K real-world free-moving camera trajectories, depth maps and detailed motion labels. It was developed by institutions including Zhejiang University and Shanghai AI Laboratory. Each sample includes a shot-level camera trajectory, corresponding RGBD images, and two types of trajectory descriptions: motion descriptions and director descriptions. Motion descriptions only describe camera motion, while director descriptions include camera motion, interactions with the scene and directorial intent. This dataset focuses on artistic free-moving trajectories, which can capture the creative vision of directors and has cinematic and artistic value.




