Based on the project (Project Page)
- Python 2.7
- TensorFlow tested on version 1.3
Run prepaired docker container
docker run -it --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=0 \
-p 2021:8888 --name human_reconstruction \
-v {data}:/data \
-v {code}:/code \
fitlab/human_reconstruction:latest bash
- Download the pre-trained models
- Prepare images by passing them through openpose: images should be cropped so that the height of the person is roughly 2/3 of the image height
- Run the demo:
python -m demo --img_paths /data/demo_imgs/good_poses.csv --load_path models/model.ckpt-55124 --out_dir /data/demo_imgs/avatars
Please see doc/train.md.
If you use this code for your research, please consider citing:
@inProceedings{liang2019shape,
title={Shape-Aware Human Pose and Shape Reconstruction Using Multi-View Images},
author = {Junbang Liang
and Ming C. Lin},
booktitle={International Conference on Computer Vision (ICCV)},
year={2019}
}
This project is derived from HMR. If you have any question, feel free to refer to the original help doc or email [email protected]. This work is supported by National Science Foundation and Elizabeth S. Iribe Professorship.