Description

 3D 人数据集包含具有特定几何图形表示的着装人类的第一个数据集。它包含约200万张图像，其中40位男性/ 40位女性执行70个动作。每个主体动作序列均从4个摄像机视图中捕获，并带有RGB，3D骨架，身体部位和衣服分割蒙版，深度图，光流和摄像机参数进行注释。

Train/Test Splits:

• Train: woman01-woman33  man01-man33
• Test:  woman34-woman40  man34-man40

Visualize data:

Download jupyter notebook. You can use the sample dataset. Full dataset can be download in the dataset website.

conda create -n 3dpeople
source activate 3dpeople
conda install matplotlib opencv pillow scipy
conda install -c conda-forge ipywidgets=7.2.1
conda install -c plotly chart-studio
jupyter nbextension enable --py widgetsnbextension
jupyter notebook vis_dataset.ipynb 

Inside notebook select the sequence to visualize: "seq_name=..." in first cell

Contact:

cvlab.iri@gmail.com

If you use this data cite:

@inproceedings{pumarola20193dpeople,
    title={{3DPeople: Modeling the Geometry of Dressed Humans}},
    author={Pumarola, Albert and Sanchez, Jordi and Choi, Gary and Sanfeliu, Alberto and Moreno-Noguer, Francesc},
    booktitle={International Conference in Computer Vision (ICCV)},
    year={2019}
}