D2HC-RMVSNet

Here is the official repository of our paper "Dense Hybrid Recurrent Multi-view Stereo Net with Dynamic Consistency Checking" (ECCV2020 Spotlight).

How to Use

Requirements

• python 3.6
• Pytorch >= 1.0.0
• CUDA >= 9.0

Install

./conda_install.sh

Training

• Download the preprocessed DTU training data (also available at Baiduyun, code: s2v2), and upzip it as the MVS_TRANING folder.
• Set dtu_data_root to your MVS_TRAINING path in env.sh Create a log folder and a model folder in wherever you like to save the training outputs. Set the log_dir and save_dir in train.sh correspondingly.
• Train: ./train.sh

Testing

• Download our pretrained model.
• Set DTU_TESTING path or TP_TESTING path for testing in env.sh.
• Set MODEL_FOLDER to ckpt and model_ckpt_index to checkpoint_list to choose pretrained model.
• Run ./eval_dtu.sh for DTU, or ./eval_tanks.sh for Tanks and Temples.

Fusion

• Run ./fusion.sh for DTU or Tanks and Temples.

Benchmark results

Results on DTU

D2HC-RMVSNet point cloud results are provided: DTU evaluation point clouds.

Evaluate the point clouds using the DTU evaluation code.

Results on Tanks and Temples

As shown on Tanks and Temples leaderboard.

Results on BlendedMVS

The corresponding point cloud is provided: BlendedMVS result.

The rest reconstructed point clouds of the validation dataset of BlendedMVS are also provided.

Citation

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

@inproceedings{yan2020dense,
  title={Dense Hybrid Recurrent Multi-view Stereo Net with Dynamic Consistency Checking},
  author={Yan, Jianfeng and Wei, Zizhuang and Yi, Hongwei and Ding, Mingyu and Zhang, Runze and Chen, Yisong and Wang, Guoping and Tai, Yu-Wing},
  booktitle={ECCV},
  year={2020}
}

Changelog

2020 December 28

Add pretrained model on BlendedMVS.