Synthetic Dataset for Free-Form Motion Control (SynFMC)
收藏资源简介:
SynFMC数据集由复旦大学、阿里巴巴达摩院和南洋理工大学联合创建,旨在为视频生成中的动态物体和相机运动控制提供高质量数据支持。该数据集包含26,000个视频,涵盖了多种物体类别和环境场景,提供了6D姿态信息、实例分割掩码、深度图等详细注释。数据集的生成过程基于规则驱动的策略,模拟了真实世界中的复杂运动模式。SynFMC数据集的应用领域包括视频生成、计算机视觉和电影制作,旨在解决现有算法在同时控制相机和物体运动方面的局限性。
The SynFMC dataset was jointly developed by Fudan University, Alibaba DAMO Academy, and Nanyang Technological University, with the goal of providing high-quality data support for dynamic object and camera motion control in video generation. This dataset contains 26,000 videos covering diverse object categories and environmental scenarios, and provides detailed annotations including 6D pose information, instance segmentation masks, depth maps, etc. The dataset generation process adopts a rule-driven strategy to simulate complex real-world motion patterns. The SynFMC dataset has applications in video generation, computer vision, and filmmaking, and aims to address the limitations of existing algorithms in simultaneously controlling both camera and object motions.

- 1Free-Form Motion Control: A Synthetic Video Generation Dataset with Controllable Camera and Object Motions复旦大学、南洋理工大学 · 2025年



