MMDepth is an open source monocular depth estimation toolbox based on PyTorch. The main branch works with PyTorch 1.6+.
Note that this repo is not a part of the OpenMMLab project. It is built on top of MMSegmentation. If there is any infringement please contact ruijiezhu@mail.ustc.edu.cn.
Since this is the first version of this repo, there are inevitably many imperfections. The code library inherits most of the functions and advantages of MMSegmentation, and adds or expands some components. We welcome you to develop and maintain this repo together!
-
Unified Benchmark
We provide a unified toolbox and benchmark for training or testing models on multiple depth datasets.
-
Modular Design
Following OpenMMLab series, we decompose the monocular depth estimation models into different components. As a result, it is easy to construct a customized monocular depth estimation model by combining different modules.
Please refer to get_started.md for installation and dataset_prepare.md for dataset preparation.
Please refer to train.md for model training and inference.md for model inference.
For detailed guides for development, please see tutorials from MMSegmentation.
Here are some implementations of SOTA models and solutions built on MMDepth, which are supported and maintained by community users. These projects demonstrate the best practices based on MMDepth for research and product development. We welcome and appreciate all the contributions to these projects. Also, we appreciate all contributions to improve MMDepth framework.
ScaleDepth: Decomposing Metric Depth Estimation into Scale Prediction and Relative Depth Estimation
Ruijie Zhu, Chuxin Wang, Ziyang Song, Li Liu, Tianzhu Zhang, Yongdong Zhang Arxiv, 2024 [Paper] [Webpage] [Code] |
|
BinsFormer: Revisiting Adaptive Bins for Monocular Depth Estimation
Zhenyu Li, Xuyang Wang, Xianming Liu, and Junjun Jiang IEEE Transactions on Image Processing, 2024 [Paper] [Code] |
Results and models are available in the model zoo.
MMDepth is an open source project maintained by the author alone (at least for now). The original intention of building this project is to provide a standardized toolbox and benchmark for monocular depth estimation. We thank the contributers of MMSegmentation for providing a great template for this project. We also thank the authors of Monocular-Depth-Estimation-Toolbox and ZoeDepth, whose code this project borrowed.
If you find this project useful in your research, please consider cite:
@misc{mmdepth2024,
title={{mmdepth}: Monocular Depth Estimation Toolbox and Benchmark},
author={Ruijie Zhu},
howpublished = {\url{https://github.com/RuijieZhu94/mmdepth}},
year={2024}
}
And if you find those SOTA models and solutions built on MMDepth useful, please also consider cite:
@article{zhu2024scaledepth,
title={ScaleDepth: Decomposing Metric Depth Estimation into Scale Prediction and Relative Depth Estimation},
author={Zhu, Ruijie and Wang, Chuxin and Song, Ziyang and Liu, Li and Zhang, Tianzhu and Zhang, Yongdong},
journal={arXiv preprint arXiv:2407.08187},
year={2024}
}
@article{li2022binsformer,
title={BinsFormer: Revisiting Adaptive Bins for Monocular Depth Estimation},
author={Li, Zhenyu and Wang, Xuyang and Liu, Xianming and Jiang, Junjun},
journal={arXiv preprint arXiv:2204.00987},
year={2022}
}
This project is released under the Apache 2.0 license.