Stereo 3D Reconstruction

This repository contains the source code for the paper Toward 3D Object Reconstruction from Stereo Images.

Important Note: The source code is in the (Stereo2Voxel/Stereo2Point) branches of the repository.

Cite this work

@article{xie2021towards,
  title={Toward 3D Object Reconstruction from Stereo Images},
  author={Xie, Haozhe and
          Tong, Xiaojun and
          Yao, Hongxun and
          Zhou, Shangchen and
          Zhang, Shengping and
          Sun, Wenxiu},
  journal={Neurocomputing},
  year={2021}
}

Datasets

We use the StereoShapeNet dataset in our experiments, which is available below:

• StereoShapeNet

Pretrained Models

The pretrained models on StereoShapeNet are available as follows:

• Stereo2Voxel for StereoShapeNet (309 MB)
• Stereo2Point for StereoShapeNet (356 MB)

Prerequisites

Clone the Code Repository

git clone https://github.com/hzxie/Stereo-3D-Reconstruction.git

Install Python Denpendencies

cd Stereo-3D-Reconstruction
pip install -r requirements.txt

Train/Test Stereo2Voxel

git checkout Stereo2Voxel

Train/Test Stereo2Point

git checkout Stereo2Point

cd extensions/chamfer_dist
python setup.py install --user

Update Settings in config.py

You need to update the file path of the datasets:

__C.DATASETS.SHAPENET.LEFT_RENDERING_PATH   = '/path/to/ShapeNetStereoRendering/%s/%s/render_%02d_l.png'
__C.DATASETS.SHAPENET.RIGHT_RENDERING_PATH  = '/path/to/ShapeNetStereoRendering/%s/%s/render_%02d_r.png'
__C.DATASETS.SHAPENET.LEFT_DISP_PATH        = '/path/to/ShapeNetStereoRendering/%s/%s/disp_%02d_l.exr'
__C.DATASETS.SHAPENET.RIGHT_DISP_PATH       = '/path/to/ShapeNetStereoRendering/%s/%s/disp_%02d_r.exr'
__C.DATASETS.SHAPENET.VOLUME_PATH           = '/path/to/ShapeNetVox32/%s/%s.mat'

Get Started

To train GRNet, you can simply use the following command:

python3 runner.py

To test GRNet, you can use the following command:

python3 runner.py --test --weights=/path/to/pretrained/model.pth

License

This project is open sourced under MIT license.