VideoLightingDirection (VLD) 数据集

VideoLightingDirection (VLD) 数据集是一个高质量合成视频数据集，由复旦大学提供，包含每帧的光照方向标注，捕捉了复杂的光照交互，如反射。该数据集使得VidCRAFT3模型能够有效地处理强烈的光照传输和反射效果。数据集的创建是为了解决真实世界视频数据集缺乏光照标注的问题，它包含了多样化的物体外观，支持VidCRAFT3在保持视觉一致性的同时对相机运动、物体运动和光照方向进行精确控制。VideoLightingDirection (VLD) Dataset is a high-quality synthetic video dataset provided by Fudan University. It contains per-frame lighting direction annotations and captures complex lighting interactions such as reflections. This dataset enables the VidCRAFT3 model to effectively handle strong light transport and reflection effects. The dataset was created to address the issue that real-world video datasets lack lighting annotations. It includes diverse object appearances and supports VidCRAFT3 in precisely controlling camera motion, object motion and lighting direction while maintaining visual consistency.

搜集汇总

背景与挑战

背景概述

VideoLightingDirection (VLD) 数据集是由复旦大学提供的高质量合成视频数据集，每帧包含光照方向标注，旨在捕捉复杂的光照交互（如反射）。该数据集解决了真实世界视频缺乏光照标注的问题，支持VidCRAFT3模型在控制相机和物体运动的同时处理强烈的光照传输和反射效果，保持视觉一致性。