MultiCamVideo-Dataset
收藏资源简介：
MultiCamVideo数据集是一个使用Unreal Engine 5渲染的多摄像头同步视频数据集。它包含了同步的多摄像头视频及其相应的相机轨迹。数据集由13,600个不同的动态场景组成，每个场景由10个摄像头捕获，共计136,000个视频。每个动态场景包括四个元素：3D环境、角色、动画和相机。使用动画驱动角色，并将其放置在3D环境中。随后，设置时间同步的相机沿预定义的轨迹移动以渲染多摄像头视频数据。数据集涵盖了多种室内外场景，如城市街道、购物中心、咖啡馆、办公室和乡村等。角色和动画从Fab和Mixamo收集，共有66种不同的人类3D模型和93种动画。相机轨迹设计多样，包括平移、旋转和弧线运动等，以及静态相机设置。数据集的统计数据为：动态场景数量13,600个，每个场景的摄像头数量10个，总视频数136,000个，压缩文件大小312G。视频配置为1280x1280分辨率，81帧数，15FPS。相机配置包括四种焦距和光圈组合。
The MultiCamVideo dataset is a synchronized multi-camera video dataset rendered using Unreal Engine 5. It contains synchronized multi-camera video footage and their corresponding camera trajectories. The dataset consists of 13,600 distinct dynamic scenes, each captured by 10 cameras, totaling 136,000 video clips. Each dynamic scene includes four core components: a 3D environment, digital characters, animations, and cameras. Characters are animated using the collected animations and placed into the 3D environment. Subsequently, time-synchronized cameras are set to move along predefined trajectories to render the multi-camera video data. The dataset covers a variety of indoor and outdoor scenarios, including urban streets, shopping malls, cafes, offices, rural areas, and more. Characters and animations are sourced from Fab and Mixamo, with a total of 66 distinct human 3D models and 93 animation clips. The camera trajectories are diversely designed, including translational, rotational, and arc movements, as well as static camera setups. The statistical specifications of the dataset are as follows: 13,600 dynamic scenes, 10 cameras per scene, 136,000 total video clips, a compressed file size of 312 GB, each video with a resolution of 1280×1280, 81 frames per clip, and a frame rate of 15 FPS. The camera configurations include four combinations of focal length and aperture.
MultiCamVideo Dataset 概述
1. 数据集简介
- 概述：MultiCamVideo Dataset是一个通过Unreal Engine 5渲染的多摄像头同步视频数据集，包含同步的多摄像头视频及其对应的摄像头轨迹。
- 应用领域：适用于摄像头控制视频生成、同步视频制作和3D/4D重建等领域。
- 构成要素：每个动态场景由四个元素组成：3D环境、角色、动画和摄像头。
2. 数据集构成
- 3D环境：收集了37个高质量的3D环境资产，涵盖室内外多种场景。
- 角色：包含66个不同的人体3D模型。
- 动画：收集了93种不同的动画，包括常见的动作如挥手、跳舞和欢呼。
- 摄像头：设计了多样化的摄像头轨迹和参数，确保运动多样性和真实世界分布。
3. 数据集统计与配置
-
数据集统计：
- 动态场景数量：13,600
- 每个场景的摄像头数量：10
- 总视频数量：136,000
- 压缩文件大小：312G
-
视频配置：
- 分辨率：1280x1280
- 帧数：81
- 帧率：15 FPS
-
摄像头配置：
- 焦距：18mm, 24mm, 35mm, 50mm
- 光圈：10.0, 5.0, 2.4
- 传感器高度：23.76mm
- 传感器宽度：23.76mm
4. 文件结构
MultiCamVideo-Dataset ├── train │ ├── f18_aperture10 │ │ ├── scene1 │ │ │ ├── videos │ │ │ │ ├── cam01.mp4 │ │ │ │ ├── cam02.mp4 │ │ │ │ ├── ... │ │ │ │ └── cam10.mp4 │ │ │ └── cameras │ │ │ └── camera_extrinsics.json │ │ ├── ... │ │ └── scene3400 │ ├── f24_aperture5 │ │ ├── scene1 │ │ ├── ... │ │ └── scene3400 │ ├── f35_aperture2.4 │ │ ├── scene1 │ │ ├── ... │ │ └── scene3400 │ └── f50_aperture2.4 │ ├── scene1 │ ├── ... │ └── scene3400 └── val └── 10basic_trajectories ├── videos │ ├── cam01.mp4 │ ├── cam02.mp4 │ ├── ... │ └── cam10.mp4 └── cameras └── camera_extrinsics.json
5. 实用脚本
-
数据提取： bash sudo apt-get install git-lfs git lfs install git clone https://huggingface.co/datasets/KwaiVGI/MultiCamVideo-Dataset cat MultiCamVideo-Dataset.part* > MultiCamVideo-Dataset.tar.gz tar -xzvf MultiCamVideo-Dataset.tar.gz
-
摄像头可视化： python python vis_cam.py
6. 引用
bibtex @misc{bai2025recammaster, title={ReCamMaster: Camera-Controlled Generative Rendering from A Single Video}, author={Jianhong Bai and Menghan Xia and Xiao Fu and Xintao Wang and Lianrui Mu and Jinwen Cao and Zuozhu Liu and Haoji Hu and Xiang Bai and Pengfei Wan and Di Zhang}, year={2025}, eprint={2503.11647}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2503.11647}, }
7. 联系方式
