Dataset proposed in "Pose2UV: Single-shot Multi-person Mesh Recovery with Deep UV Prior"
[ProjectPage] [paper] @ TIP2022.

Samples of 3DMPB Dataset

3DMPB is a multi-person dataset in the outdoor sport field with human interaction occlusion and image truncation. This dataset provides annotations including bounding-box, human 2D pose, SMPL model annotations, instance mask and camera parameters.

Note: Not all the instances in this dataset are annotated in consideration of some inaccurate annotations or wrong relative occlusion.

[Baidu Netdisk] [Google Drive]

We also provide full videos for 3DMPB dataset.

[3DMPB-video]

• python3
• numpy
• pytorch
• pyrender

Please download the dataset and extract under ./3DMPB. Due to the licenses, please download SMPL model file here. The folder structure is shown as follows:

|-- 3DMPB
`-- |-- images
    |   |-- 000000.jpg
    |   |-- 000001.jpg
    |   |-- 000002.jpg
    |   |-- ...
    |-- masks
    |   |-- 000000_00.jpg
    |   |-- 000000_01.jpg
    |   |-- 000000_02.jpg
    |   |-- 000001_00.jpg
    |   |-- 000002_00.jpg
    |   |-- ...
    |-- annot.json
|-- data
`-- |-- SMPL_NEUTRAL.pkl    

python vis_dataset.py --dataset_dir 3DMPB  --output_dir output 

python vis_dataset.py --dataset_dir 3DMPB  --output_dir output --vis_smpl True

If you find this dataset useful for your research, please consider citing the paper.

@article{huang2022pose2uv,
  title={Pose2UV: Single-shot Multi-person Mesh Recovery with Deep UV Prior},
  author={Huang, Buzhen and Zhang, Tianshu and Wang, Yangang},
  journal={IEEE Transactions on Image Processing},
  year={2022},
  volume={31},
  pages={4679-4692}
}