Skip to content

Latest commit

 

History

History
88 lines (80 loc) · 3.47 KB

README.md

File metadata and controls

88 lines (80 loc) · 3.47 KB
Raw

StereoPIFu: Depth Aware Clothed Human Digitization via Stereo Vision

| Project Page | Paper |

This repository contains a pytorch implementation of "StereoPIFu: Depth Aware Clothed Human Digitization via Stereo Vision (CVPR 2021)".
Authors: Yang Hong, Juyong Zhang, Boyi Jiang, Yudong Guo, Ligang Liu and Hujun Bao.

Requirements

  • Python 3
  • Pytorch (<=1.4.0, some compatibility issues may occur in higher versions of pytorch)
  • tqdm
  • opencv-python
  • scikit-image
  • openmesh

for building evaluation data

  • pybind11,we recommend "pip install pybind11[global]" for installation.
  • gcc
  • cmake

Run the following code to install all pip packages:

pip install -r requirements.txt

Building Evaluation Data

Preliminary

Run the following script to compile & generate the relevant python module, which is used to render left/right color/depth/mask images from the textured/colored mesh.

cd GenEvalData
bash build.sh
cd ..

Usage

#demo, for textured mesh
python GenEvalData.py \
--tex_mesh_path="TempData/SampleData/rp_dennis_posed_004_100k.obj" \
--tex_img_path="TempData/SampleData/rp_dennis_posed_004_dif_2k.jpg" \
--save_dir="./TempData/TexMesh" \
--save_postfix="tex"
#demo, for colored mesh
python GenEvalData.py \
--color_mesh_path="TempData/SampleData/normalized_mesh_0089.off" \
--save_dir="./TempData/ColorMesh" \
--save_postfix="color"

These samples are from renderpeople and BUFF dataset.
Note: the mesh used for rendering needs to be located in a specific bounding box.

Inference

Preliminary

  • Run the following script to compile & generate deformable convolution from AANet. 
    cd AANetPlusFeature/deform_conv
bash build.sh
cd ../..
  • Download the trained model and mv to the "Models" folder.
  • Generate evalution data with aboved "Building Evaluation Data", or capture real data by ZED Camera (we test on ZED camera v1).
    Note: rectifying left/right images is required before using ZED camera.

Demo

bash eval.sh

The reconsturction result will be saved to "Results" folder.
Note: At least 10GB GPU memory is recommended to run StereoPIFu model.

Citation

@inproceedings{yang2021stereopifu,
  author    = {Yang Hong and Juyong Zhang and Boyi Jiang and Yudong Guo and Ligang Liu and Hujun Bao},
  title     = {StereoPIFu: Depth Aware Clothed Human Digitization via Stereo Vision},
  booktitle = {{IEEE/CVF} Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2021}
}

Contact

If you have questions, please contact [email protected].