BlendedMVS is a large-scale MVS dataset for generalized multi-view stereo networks. The dataset contains over 17k MVS training samples covering a variety of scenes, including architectures, sculptures and small objects. If you find the dataset useful for your research, please cite:
@article{yao2019blendedmvs,
title={BlendedMVS: A Large-scale Dataset for Generalized Multi-view Stereo Networks},
author={Yao, Yao and Luo, Zixin and Li, Shiwei and Zhang, Jingyang and Ren, Yufan and Zhou, Lei and Fang, Tian and Quan, Long},
journal={arXiv:1911.10127 [cs.CV]},
year={2019}
}
To visualize each scene via Altizure online platform, please go to https://www.altizure.com/project-model?pid=PID, where PID is the unique project ID (e.g., 5bfe5ae0fe0ea555e6a969ca). All PID are listed in the all_list.txt file.
BlendedMVS is licensed under a Creative Commons Attribution 4.0 International License!!!
We provide both high-resolution and low-resolution training data in BlendedMVS dataset.
- High-resolution set: Image and depth map resolutions are set to 2048 x 1536.
- Low-resolution set: Image and depth map resolutions are set to 768 x 576.
Currently the low-resolution set is provided for MVSNet training, the high-resolution set is provided upon request.
BlendedMVS dataset adopts MVSNet input format. The dataset structure is listed below:
BlendedMVS
├── all_list.txt
├── training_list.txt
├── validation_list.txt
├── PID0
│ ├── blended_images
│ │ ├── 00000000.jpg
│ │ ├── 00000001.jpg
│ │ └── ...
│ ├── cams
│ │ ├── 00000000_cam.txt
│ │ ├── 00000001_cam.txt
│ │ └── ...
│ ├── rendered_depth_maps
│ │ ├── 00000000.pfm
│ │ ├── 00000001.pfm
│ │ └── ...
│ └── pair.txt
├── PID1
├── ...
└── PID112
PID here is the unique project ID listed in the all_list.txt file. For file formats of other training inputs, users may refer to MVSNet for more details.
- Training and validation codes.