ReCamMaster
收藏资源简介:
ReCamMaster数据集是由浙江大学和快手科技共同创建的高质量多相机同步视频数据集。该数据集使用虚幻引擎5构建,包含13.6万段真实感视频,涵盖13.6千个动态场景,拍摄于40个高质量的3D环境中,拥有122千个不同的相机轨迹。特别地,该数据集经过精心策划,模拟真实世界的拍摄特性,有利于模型对野外视频的泛化。数据集旨在为视频生成、4D重建等相关领域的研究提供帮助。
The ReCamMaster dataset is a high-quality multi-camera synchronized video dataset jointly created by Zhejiang University and Kuaishou Technology. Constructed using Unreal Engine 5, it contains 136,000 photorealistic video clips, covering 13,600 dynamic scenes across 40 high-quality 3D environments, and includes 122,000 unique camera trajectories. Notably, this dataset has been meticulously curated to simulate real-world filming characteristics, which facilitates the generalization of models to in-the-wild videos. It is designed to support research in relevant fields such as video generation and 4D reconstruction.

- 1ReCamMaster: Camera-Controlled Generative Rendering from A Single Video浙江大学, 快手科技, 香港中文大学, 华中科技大学 · 2025年



