A-TVSNet

Code for "A-TVSNet: Aggregated Two-View Stereo Network for Multi-View Stereo Depth Estimation"

If you find this code useful in your research, please consider citing our paper:

@article{dai2020tvsnet,
  title={A-TVSNet: Aggregated Two-View Stereo Network for Multi-View Stereo Depth Estimation},
  author={Dai, Sizhang and Huang, Weibing},
  journal={arXiv preprint arXiv:2003.00711},
  year={2020}
}

Enviroment

• Ubuntu 16.04
• Python 2.7
• Cuda 9.0 and cudnn 7.0
• Tensorflow 1.5
• other dependencies in requirements.txt

Run Examples

• Extract model.zip to model folder
• Run example

  cd atvsnet
  python example.py

Reproduce ETH3D Pointcloud Results

• Download preprocess ETH3D dataset from MVSNet and extract to data folder
• Build fusible

  cd fusible
  mkdir build
  cd build
  cmake ..
  make

• Produce point cloud

  cd atvsnet
  ./reproduce_pc.sh