Indoor Depth Completion with Boundary Consistency and Self-Attention

Official pytorch implementation of "Indoor Depth Completion with Boundary Consistency and Self-Attention. Huang et al. RLQ@ICCV 2019." arxiv

In "Indoor Depth Completion with Boundary Consistency and Self-Attention. Huang et al. RLQ@ICCV 2019.", we design a neural network which utilizes self-attention mechanism and boundary consistency concept to improving completion depth maps. Our work enhances the depth map quality and structure, which outperforms previous state-of-the-art depth completion work on Matterport3D dataset.

Implementation details and experiment results can be seen in the paper.

2022-09-26: We have released our training and testing dataset due to the missing dataset issue. Detailed instructions for downloading the dataset are described in the dataset section

Environment Setup

On x86_64 GNU/Linux machine using Python 3.6.7

git clone [email protected]:patrickwu2/Depth-Completion.git
cd Depth-Completion
pip3 install -r requirements.txt

Training / Testing

Please see train_test

Visualization / Evaluation

Please see vis_eval

Authors

Yu-Kai Huang kaikai4n [email protected]

Tsung-Han Wu tsunghan-wu [email protected]

Please cite our papers if you use this repo in your research:

@inproceedings{huang2019indoor,
  title={Indoor depth completion with boundary consistency and self-attention},
  author={Huang, Yu-Kai and Wu, Tsung-Han and Liu, Yueh-Cheng and Hsu, Winston H},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops},
  pages={0--0},
  year={2019}
}

Acknowledgement

This work was supported in part by the Ministry of Science and Technology, Taiwan, under Grant MOST 108-2634-F-002-004, FIH Mobile Limited, and Qualcomm Technologies, Inc., under Grant NAT-410477. We are grateful to the National Center for High-performance Computing.