3D Human Pose and Shape Estimation via HybrIK-Transformer

This is the implementation of this paper: https://arxiv.org/abs/2302.04774

If you use this code in any context, please cite the following paper:

@misc{oreshkin20233d,
      title={{3D} Human Pose and Shape Estimation via {HybrIK-Transformer}}, 
      author={Boris N. Oreshkin},
      year={2023},
      eprint={2302.04774},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Based on this original code: https://github.com/Jeff-sjtu/HybrIK

Register to get SMPL model

here: https://smpl.is.tue.mpg.de/register.php https://smplify.is.tue.mpg.de/register.php

Create workspace and clone this repository

mkdir workspace
cd workspace
git clone git@github.com:boreshkinai/hybrik-transformer.git

Build docker image and launch container

cd hybrik-transformer
docker build -f Dockerfile -t hybrik_transformer:$USER .

nvidia-docker run -p 8888:8888 -p 6000-6010:6000-6010 -v ~/workspace/hybrik-transformer:/workspace/hybrik-transformer -t -d --shm-size="16g" --name hybrik_transformer_$USER hybrik_transformer:$USER

Enter docker container and download data locally

docker exec -i -t hybrik_transformer_$USER  /bin/bash 
bash scripts/download_data_gcp.sh

Launch training session

docker exec -i -t hybrik_transformer_$USER  /bin/bash 
cd HybrIK
./scripts/train_smpl.sh hybrik_transformer ./configs/hybrik_transformer_smpl24.yaml

MODEL ZOO

Backbone	Training Data	PA-MPJPE (3DPW)	MPJPE (3DPW)	PA-MPJPE (Human3.6M)	MPJPE (Human3.6M)	Download	Config
ResNet-34	w/o 3DPW					model	cfg
ResNet-34	w/ 3DPW	46.0	74.9	34.6	50.2	model	cfg
HRNet-W48	w/o 3DPW	43.4	73.6	29.8	48.8	model	cfg
HRNet-W48	w/ 3DPW	42.3	71.6	29.5	47.5	model	cfg