Hao Zhu, Xinxin Zuo, Sen Wang, Xun Cao, Ruigang Yang     CVPR 2019 Oral

[project page]     [paper]

From green bounded frame: source image --> initial guess --> joint deform --> anchor deform --> vertex deform

The project is tested on ubuntu 16.04 with python 2.7, PyTorch 1.0. We recomend using Anaconda to create a new enviroment:

conda create -n py27-hmd python=2.7
conda activate py27-hmd

Install dependecies:

sudo apt-get install libsuitesparse-dev
pip install -r requirements.txt

The installation of OpenDR is unstable now, we recommend using a old stable version:

pip install pip==8.1.1
pip install opendr==0.76
pip install --upgrade pip

Refer to the guide to install the PyTorch 1.0.

Download the pretrained model:

cd demo
chmod +x download_pretrained_model.sh
./download_pretrained_model.sh

Run the demo:

python demo.py --ind 2 # or 477, 2040, 2726

The results will be saved in the folder "demo/results/" by default. Run "python demo.py -h" for more usages.

This repository merely contains 4 samples for demo. To run the full test data, download the test set from Google Drive or Baidu Cloud(extracting code:0ch3). Extract the test set and change the "dataset_path" in "conf.ini" to the extracted location. The range of test data number is [0-4624]. You can also follow the instructions in the "Data preparation" part to generate testing data together with training data.

In the generation of the dataset, we predicted the initial mesh using HMR and saved it as "/para/*.json" files. To test on images beyond the dataset, you have to run HMR to get the initial mesh firstly. The demo for images beyond the test set will be added in the near future.

Please see the datasets/data.md.

If you find this project useful for your research, please consider citing:

@inproceedings{zhu2019detailed,
  title={Detailed human shape estimation from a single image by hierarchical mesh deformation},
  author={Zhu, Hao and Zuo, Xinxin and Wang, Sen and Cao, Xun and Yang, Ruigang},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2019}
}