DepthVLM-Bench

DepthVLM-Bench is a unified indoor-outdoor metric depth estimation benchmark designed for vision-language models (VLMs). The benchmark provides diverse indoor and outdoor scenes with metric depth annotations in a unified VLM-compatible format, enabling large multimodal models to jointly learn dense geometry prediction and multimodal understanding.

Features

• Unified indoor and outdoor metric depth estimation
• VLM-compatible data format
• Dense depth supervision for multimodal foundation models
• Designed for scalable multimodal training

Paper

Unlocking Dense Metric Depth Estimation in VLMs

Usage

Please refer to the official repository for:

• Data preprocessing
• Evaluation scripts
• Visualization examples

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

Citation

@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}
}