Home

Awesome

PI-Net: Pose Interacting Network for Multi-Person Monocular 3D Pose Estimation

This is the code for the paper Wen Guo, Enric Corona, Francesc Moreno-Noguer, Xavier Alameda-Pineda, PI-Net: Pose Interacting Network for Multi-Person Monocular 3D Pose Estimation, in WACV2021.

Dependencies

Our code is tested on CUDA9, Python3.6, Pytorch1.3.0 MATLAB is needed for evaluating the 3DPCK errors.

Directory

ROOT
|-- data
|-- model
|-- utils 
`-- output
    |-- log
    |-- result 
    |-- snapshot 
        `-- snapshot_24.pth.tar
    |-- tensorboard_log 
    `-- vis 
data
|-- MuCo
    |-- MuCo.py
    `-- data
        |-- augmented_set
        |-- annotations
            |-- MuCo-3DHP_with_posenent_result_filter.json
            |-- MuCo_id2pairId.json
            `-- split_gt.py
|--MuPoTS_skeleton
    |-- MuPoTS_skeleton.py
    |-- bbox_root
    `-- data
        |-- MultiPersonTestSet
        |-- eval
        |-- MuPoTS-3D_with_posenent_result.json
        |-- MuPoTS-3D_id2pairId.json 
        `-- split_gt.py  

Preparing data

Training and testing

Citing

If you use our code, please cite our work @inproceedings{guo2021pi, title={PI-Net: Pose Interacting Network for Multi-Person Monocular 3D Pose Estimation}, author={Guo, Wen and Corona, Enric and Moreno-Noguer, Francesc and Alameda-Pineda, Xavier}, booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision}, pages={2796--2806}, year={2021} }

Acknowledgments

The overall code framework is adapted from 3DMPPE and Torchseg. The predictor model code is adapted from SeeWoLook.

Licence

MIT