MVHumanNet++
收藏资源简介:
MVHumanNet++是一个大规模的多视角日常着装人类捕捉数据集,包含4500个人类身份、9000套日常服装、60000个动作序列和6.45亿个带有广泛注释的数据。作为MVHumanNet的扩展,MVHumanNet++新增了处理过的法线图和深度图,显著扩展了其在高级人类中心研究中的适用性和实用性。此外,还提供了增强的分割和更好的SMPLX估计。
MVHumanNet++ is a large-scale multi-view daily dressing human capture dataset, encompassing 4,500 human identities, 9,000 sets of daily attire, 60,000 action sequences, and 645 million data points with extensive annotations. As an extension of MVHumanNet, MVHumanNet++ introduces processed normal maps and depth maps, significantly enhancing its applicability and practicality in advanced human-centric research. Additionally, it provides improved segmentation and better SMPLX estimation.
MVHumanNet++ 数据集概述
数据集简介
- 名称: MVHumanNet++
- 类型: 多视角日常着装人体捕捉数据集
- 规模:
- 4,500个不同人体身份
- 9,000套日常服装
- 60,000个动作序列
- 645百万帧图像数据
- 扩展内容: 新增法线贴图和深度图处理,增强分割和SMPLX估计
主要标注内容
- 人体掩码
- 相机参数
- 2D/3D关键点
- SMPL/SMPLX参数
- 文本描述
- 法线贴图
- 深度图
数据目录结构
|-- main_data |-- train |-- [ID] (如100831) |-- images # 1024x1024分辨率 |-- cam_00 # 16个固定视角相机 |-- [帧号].jpg (每25帧采样) |-- A-pose专用图像 |-- masks # 1024x1024分辨率 |-- cameras # 相机参数 |-- openpose # 姿态数据 |-- smplx_params # SMPLX参数 |-- test # 测试数据(部分掩码未手动过滤) |-- absdepth # 绝对深度数据 |-- [ID] |-- depths |-- [相机视角] |-- [帧号].exr |-- normal # 法线贴图 |-- [ID] |-- normal |-- [相机视角] |-- [帧号].jpg
使用条款
- 授权: 仅限非商业研究和教育用途
- 获取方式:
- 阅读使用条款
- 填写申请表格(https://docs.google.com/forms/d/e/1FAIpQLSeqvmeuJzUdmdU-JQ3ocu-o58Uo1K6xfxaeNOFn-AwYOxCgzQ/viewform)
相关资源
- 项目主页: https://kevinlee09.github.io/research/MVHumanNet++/
- 论文: https://arxiv.org/abs/2505.01838
- 可视化工具: vis_smplx.py
引用格式
bibtex @inproceedings{xiong2024mvhumannet, title={MVHumanNet: A Large-scale Dataset of Multi-view Daily Dressing Human Captures}, author={Xiong, Zhangyang and Li, Chenghong and Liu, Kenkun and Liao, Hongjie and Hu, Jianqiao and Zhu, Junyi and Ning, Shuliang and Qiu, Lingteng and Wang, Chongjie and Wang, Shijie and others}, booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}, year={2024} } @misc{li2025mvhumannetlargescaledatasetmultiview, title={MVHumanNet++: A Large-scale Dataset of Multi-view Daily Dressing Human Captures with Richer Annotations for 3D Human Digitization}, author={Chenghong Li and Hongjie Liao and Yihao Zhi and Xihe Yang and Zhengwentai Sun and Jiahao Chang and Shuguang Cui and Xiaoguang Han}, year={2025}, eprint={2505.01838}, archivePrefix={arXiv}, primaryClass={cs.CV} }




