Eigenpose_release

This repository is an official Pytorch implementation of the paper "Eigenpose: Occlusion-robust 3D Human Mesh Reconstruction".
Mi-Gyeong Gwon, Gi-Mun Um, Won-Sik Cheong, Wonjun Kim (Corresponding Author)

IEEE Transactions on Image Processing (TIP)

Installation

Environment Setting

Please set the environment and datasets by following the guidance of ROMP.

Model Checkpoints (only for evaluation)

Create checkpoints directory and place the model checkpoint file in it.

Run

Evaluation

$ python -m romp.test --configs_yml=configs/eval_3dpw_test_resnet.yml
$ python -m romp.test --configs_yml=configs/eval_ochuman_resnet_test.yml
$ python -m romp.test --configs_yml=configs/eval_crowdpose_test.yml
$ python -m romp.test --configs_yml=configs/eval_oh50k_test.yml

※ You can change the subset of 3DPW (3DPW-PC, 3DPW-OC) by changing the eval_dataset setting in the file configs/eval_3dpw_test_resnet.yml.

Results

Results of 3D human mesh reconstruction by the proposed method on 3DPW (1st row) and COCO (2nd and 3rd rows) datasets.

Citation

If you find our work useful for your project, please consider citing the following paper.

@misc{gwon2025eigenpose,
      title={Eigenpose: Occlusion-Robust 3D Human Mesh Reconstruction}, 
      author={Gwon, Mi-Gyeong and Um, Gi-Mun and Cheong, Won-Sik and Kim, Wonjun},
      year={2025},
      volume={34},
      pages={2379--2391},
      journal={IEEE Transactions on Image Processing},
      publisher={IEEE}
      doi={10.1109/TIP.2025.3559788}
}