Update 2026-05-18 (v1.0): Initial release

DepthVLM-4B

DepthVLM serves as a unified foundation model for both low-level dense geometry prediction and high-level multimodal understanding, while achieving substantially faster inference compared with existing VLM-based approaches such as DepthLM and Youtu-VL.

Highlights

• Native dense metric depth estimation in VLMs
• Unified multimodal understanding and geometry prediction
• Full-resolution depth prediction with efficient inference
• Supports both indoor and outdoor metric depth estimation
• Improved 3D spatial reasoning capability

Paper

Unlocking Dense Metric Depth Estimation in VLMs

Usage

Please refer to the official repository for detailed instructions on:

• Data preprocessing
• Training
• Evaluation
• Inference and visualization

Repository: https://github.com/hanxunyu/DepthVLM

Citation

If you find this work useful, please cite:

@article{yu2026unlocking,
  title={Unlocking Dense Metric Depth Estimation in VLMs},
  author={Hanxun Yu and Xuan Qu and Yuxin Wang and Jianke Zhu and Lei Ke},
  journal={arXiv preprint arXiv:2605.15876},
  year={2026}
}