SynCamVideo-Dataset Multi-camera Synchronized Video Dataset

SynCamVideo-Dataset is a multi-camera synchronized video dataset generated by a research team from Zhejiang University, Kuaishou Technology, Tsinghua University, and the Chinese University of Hong Kong in 2024 using Unreal Engine 5 rendering. The related paper results are "SynCamMaster: Synchronizing Multi-Camera Video Generation from Diverse Viewpoints".

Contains 1k different scenes, each captured by 36 cameras, generating a total of 36k videos. The dataset features 50 different animals as "main objects" and uses 20 different locations from Poly Haven as backgrounds. The cameras in each scene are placed on a hemispherical surface at a distance of 3.5 meters to 9 meters from the center of the scene. To ensure that the rendered video has minimal field offset from the real-world video, the camera's elevation angle is constrained to between 0° and 45°, and the azimuth angle is constrained to between 0° and 360°. Each camera is randomly sampled under these constraints, rather than using the same set of camera positions in all scenes.

The SynCamVideo dataset can be used to train multi-camera synchronized video generation models, inspiring applications in areas such as filmmaking and multi-view data generation. The file structure of the dataset contains training and validation sets, with the video and corresponding camera extrinsic parameters of each scene stored separately.