MVHumanNet: A Large-scale Dataset of Multi-view Daily
Dressing Human Captures

Project Page | Paper (Arxiv) | Dataset

by Zhangyang Xiong#, Chenghong Li#, Kenkun Liu#, Hongjie Liao, Jianqiao Hu, Junyi Zhu, Shuliang Ning, Lingteng Qiu, Chongjie Wang, Shijie Wang, Shuguang Cui and Xiaoguang Han* from GAP-Lab.

Introduction

MVHumanNet contains 4,500 human identities, 9,000 daily outfits, 60,000 motion sequences, 645 million with extensive annotations, including human masks, camera parameters , 2D and 3D keypoints, SMPL/SMPLX parameters, and corresponding textual descriptions.

Updates （MVHumanNet_Part1 is available now!）

2024.05.07: MVHumanNet_Part1 is released! 🔥🔥🔥🔥🔥🔥🔥

MVHumanNet_Part1 contains about 2500 IDs and 4800 outfits. We provide links to download the MVHumanNet_Part1. Please fill out the form to get the download links.

The rest data will be available soon.

2023.12.20 Samples of MVHumanNet are available now！！！

These samples contain 100 outfits, with 6+1 motions sequences for each. We provide a link to download the samples. Please fill out this form to get the download link.

Folder structure

|-- ROOT
    |-- outfits_ID # 100001
        |-- images    # Considering the limitation of storage space, we scaled the image to half the original size and masked some background.
            |-- camera_name
                |-- images  
            |-- camera_name
                |-- images
            ....
        |-- fmask   # corresponding masks.
            |-- camera_name
                |-- mask images 
            |-- camera_name
                |-- mask images 
            ....
        |-- annots # 2D image annotations by openpose.
            |-- camera_name
                |-- annotations  # json files
            |-- camera_name
                |-- annotations  # json files
            ....
        |-- openpose
            |-- camera_name
                |-- 2D keypoints  # json files
            |-- camera_name
                |-- 2D keypoints  # json files
            ....
        |-- smpl_param  #  optimizes from multi-view images
            |-- PKL files
        |-- smplx  #  optimizes from multi-view images
            |-- 3D keypoints
                |-- json files # 3D keypoints
            |-- smpl
                |-- json files 
            |-- smplx_mesh  
                |-- obj files  # smplx meshs
        |-- camera_extrinsics.json   # extrinsics of all cameras
        |-- camera_intrinsics.json   # intrinsics of all cameras
        |-- camera_scale.pkl   #

If you find our work useful in your research, please consider citing:

@article{xiong2023mvhumannet,
    title     = MVHumanNet: A Large-scale Dataset of Multi-view Daily Dressing Human Captures },
    author    = {Zhangyang Xiong, Chenghong Li, Kenkun Liu, Hongjie Liao, Jianqiao HU, Junyi Zhu, Shuliang Ning, Lingteng Qiu, Chongjie Wang, Shijie Wang, Shuguang Cui and Xiaoguang Han
},
    journal={arXiv preprint},
    year={2023}
}

License

The data is released under the MVHumanNet Terms of Use, and the code is released under the Attribution-NonCommercial 4.0 International License.

GAP-LAB-CUHK-SZ / MVHumanNet