Awesome

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)".<br/> 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 

<span id="Building_Evaluation"></span>

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.<br/> 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). <br/>Note: rectifying left/right images is required before using ZED camera.

Demo

bash eval.sh

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

Contact

If you have questions, please contact hymath@mail.ustc.edu.cn.