🔰 Project Page, 📑 Paper
Yongtao Ge1,2, Guangkai Xu1, Zhiyue Zhao1, Libo Sun2, Zheng Huang1, Yanlong Sun3, Hao Chen1, Chunhua Shen1
1Zhejiang University, 2The University of Adelaide, 3Tsinghua University
This toolbox streamlines the use and evaluation for state-of-the-art discriminative and generative geometry estimation models, which can be served as foundation models for various downstream 3D reconstruction applications, including:
- Metric3D-V2
- UniDepth
- Depth-Anything-V2
- Depth-Anything
- DSINE
- Marigold
- DMP
- Genpercept
- Geowizard
- DepthFM
We would also support multi-view geometry models in the near future:
pip install -r requirements.txt
pip install -e . -v
# inference Marigold
sh scripts/run_marigold.sh
# inference Metric3D
sh scripts/run_metric3d.sh
# inference Depth-Anything
sh scripts/run_depthanything.sh
# inference GenPercept
sh scripts/run_genpercept.sh
# inference DSINE
sh scripts/run_dsine.sh
Stay tuned, comming soon.
If you find the toolbox useful for your project, please cite our paper:
@article{ge2024geobench,
title={GeoBench: Benchmarking and Analyzing Monocular Geometry Estimation Models},
author={Ge, Yongtao and Xu, Guangkai, and Zhao, Zhiyue and Huang, zheng and Sun, libo and Sun, Yanlong and Chen, Hao and Shen, Chunhua},
journal={arXiv preprint arXiv:2406.12671},
year={2024}
}
GeoBench is released under the Apache 2.0 license.