MonoRelief V2: Leveraging Real Data for High-Fidelity Monocular Relief Recovery

Yu-Wei Zhang^1*† · Tongju Han¹ ·Lipeng Gao¹
Mingqiang Wei² · Hui Liu³ · Changbao Li¹ · Caiming Zhang⁴

¹QLU   ²NUAA   ³SDUFE   ⁴SDU
†project lead *corresponding author

This paper presents MonoRelief V2, an end-to-end model designed for directly recovering 2.5D reliefs from single images under complex material and illumination variations. In contrast to its predecessor, MonoRelief V1, which was solely trained on synthetic data, MonoRelief V2 incorporates real data to achieve improved robustness, accuracy and efffciency. To overcome the challenge of acquiring large-scale real-world dataset, we generate approximately 15,000 pseudoreal images using a text-to-image generative model, and derive corresponding depth pseudo-labels through fusion of depth and normal predictions. Furthermore, we construct a small-scale real-world dataset (800 samples) via multi-view reconstruction and detail reffnement. MonoRelief V2 is then progressively trained on the pseudo-real and real-world datasets. Comprehensive experiments demonstrate its state-of-the-art performance both in depth and normal predictions, highlighting its strong potential for a range of downstream applications.

Notices

Please note that MonoRelief is designed to process relief images as input, rather than natural scene images (such as portraits of people, animals, etc.). If you wish to create relief models from realistic natural scenes, we recommend first using image editing tools like ChatGPT-Image or NanoBanana to convert your images into a relief style, and subsequently feeding them into MonoRelief for depth recovery.

News

• 2025-08-28: updata readme.
• 2025-09-05: Paper and codes are released.
• 2026-06-01 : We have upgraded MonoRelief to Version 3. If you would like to test it, please contact us via zhangyuwei_scott@126.com.

Acknowledgements

We gratefully acknowledge the following open-source projects that our work builds upon:

• Depth-Anything-V2 (repo).

Citation

If you find this project useful, please consider citing:

@article{gao2025monorelief,
  title={MonoRelief: Recovering 2.5 D Relief from a Single Image},
  author={Gao, Lipeng and Zhang, Yu-Wei and Wei, Mingqiang and Liu, Hui and Chen, Yanzhao and Qiu, Huadong and Zhang, Caiming},
  journal={IEEE Transactions on Visualization and Computer Graphics},
  year={2025},
  publisher={IEEE}
}

@article{zhang2025monorelief,
  title={MonoRelief V2: Leveraging Real Data for High-Fidelity Monocular Relief Recovery},
  author={Zhang, Yu-Wei and Han, Tongju and Gao, Lipeng and Wei, Mingqiang and Liu, Hui and Li, Changbao and Zhang, Caiming},
  journal={arXiv preprint arXiv:2508.19555},
  year={2025}
}