Awesome

U-MVS

Official code for ICCV paper "Digging into Uncertainty in Self-supervised Multi-view Stereo" [Paper] [Arxiv]

Log

2022-01-12

• The evaluation code of U-MVS(CasMVSNet) is released.
• The training code will be uploaded later.

2022-01-02

• The training code has been uploaded.

2021-12-24

• The evaulation code of U-MVS(MVSNet) is released.
• The training code will be uploaded in a few days.
• A toy example for understanding the depth2flow module is provided in toy_example_depth2flow.

Citation

If you find this work is helpful to your work, please cite:

@inproceedings{xu2021digging,
  title={Digging into Uncertainty in Self-supervised Multi-view Stereo},
  author={Xu, Hongbin and Zhou, Zhipeng and Wang, Yali and Kang, Wenxiong and Sun, Baigui and Li, Hao and Qiao, Yu},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={6078--6087},
  year={2021}
}

Acknowledgement

The baseline code of this repository is based on JDACS. We also acknowledge the code of arflow for their great work in unsupervised flow estimation^[1], which is used as the backbone of our RGB2Flow module. Furthermore, we thank for the Tensorflow implementation in dl-uncertainty for aleatoric and epistemic uncertainty estimation^[2].

Reference

[1] L Liu, J Zhang, and etc, "Learning by Analogy: Reliable Supervision from Transformations for Unsupervised Optical Flow Estimation", CVPR 2020

[2] A Kendall, Y Gal, "What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?", NIPS 2017