Update README.md
Browse files
README.md
CHANGED
|
@@ -12,4 +12,41 @@ tags:
|
|
| 12 |
|
| 13 |
paper:
|
| 14 |
- arxiv: 2605.15876
|
| 15 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
paper:
|
| 14 |
- arxiv: 2605.15876
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# DepthVLM Benchmark
|
| 18 |
+
|
| 19 |
+
DepthVLM Benchmark is a unified indoor-outdoor metric depth estimation benchmark designed for vision-language models (VLMs).
|
| 20 |
+
|
| 21 |
+
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.
|
| 22 |
+
|
| 23 |
+
## Features
|
| 24 |
+
|
| 25 |
+
- Unified indoor and outdoor metric depth estimation
|
| 26 |
+
- VLM-compatible data format
|
| 27 |
+
- Dense depth supervision for multimodal foundation models
|
| 28 |
+
- Supports both geometry prediction and spatial reasoning tasks
|
| 29 |
+
- Designed for scalable multimodal training
|
| 30 |
+
|
| 31 |
+
## Paper
|
| 32 |
+
|
| 33 |
+
[Unlocking Dense Metric Depth Estimation in VLMs](https://arxiv.org/abs/2605.15876)
|
| 34 |
+
|
| 35 |
+
## Usage
|
| 36 |
+
|
| 37 |
+
Please refer to [the official repository](https://github.com/hanxunyu/DepthVLM) for:
|
| 38 |
+
|
| 39 |
+
- Data preprocessing
|
| 40 |
+
- Training instructions
|
| 41 |
+
- Evaluation scripts
|
| 42 |
+
- Visualization examples
|
| 43 |
+
|
| 44 |
+
## Citation
|
| 45 |
+
|
| 46 |
+
```bibtex id="83r6sk"
|
| 47 |
+
@article{yu2026unlocking,
|
| 48 |
+
title={Unlocking Dense Metric Depth Estimation in VLMs},
|
| 49 |
+
author={Hanxun Yu and Xuan Qu and Yuxin Wang and Jianke Zhu and Lei Ke},
|
| 50 |
+
journal={arXiv preprint arXiv:2605.15876},
|
| 51 |
+
year={2026}
|
| 52 |
+
}
|