G2VLM-Qwen2-VL-2B

G2VLM-Qwen2-VL-2B

Geometry Grounded Vision Language Model with Unified 3D Reconstruction and Spatial Reasoning

<p align="left"> <img src="https://huggingface.co/InternRobotics/G2VLM-2B-MoT/resolve/main/assets/icon.png" alt="G2VLM" width="200"/> </p> <p align="left"> <a href="https://gordonhu608.github.io/g2vlm.github.io/"> <img src="https://img.shields.io/badge/G2VLM-Website-0A66C2?logo=safari&logoColor=white" style="display: inline-block; vertical-align: middle;" alt="G2VLM Website" /> </a> <a href="https://arxiv.org/abs/2511.21688"> <img src="https://img.shields.io/badge/G2VLM-Paper-red?logo=arxiv&logoColor=red" style="display: inline-block; vertical-align: middle;" alt="G2VLM Paper on arXiv" /> </a> <a href="https://github.com/InternRobotics/G2VLM" target="_blank" style="margin: 2px;"> <img alt="Github" src="https://img.shields.io/badge/G2VLM-Codebase-536af5?color=536af5&logo=github" style="display: inline-block; vertical-align: middle;" alt="G2VLM Codebase" /> </a> </p>

We present <b>G²VLM</b>, a geometry grounded vision-language model proficient in both spatial 3D reconstruction and spatial understanding tasks. For spatial reasoning questions, G²VLM can natively predict 3D geometry and employ interleaved reasoning for an answer.

This repository hosts the base model weights <b>BEFORE</b> the training of <b>G²VLM</b>, which is technically the same as Qwen2-VL-2B. Here we format it so it's easier for users to reproduce our trainings. For installation, usage instructions, and further documentation, please visit our GitHub repository.

<p align="left"><img src="https://huggingface.co/InternRobotics/G2VLM-2B-MoT/resolve/main/assets/teaser.png" width="100%"></p>

🧠 Method

G²VLM is a unified model that integrates both a geometric perception expert for 3D reconstruction and a semantic perception expert for multimodal understanding and spatial reasoning tasks. All tokens can do shared multi-modal self attention in each transformer block.

<p align="left"><img src="https://huggingface.co/InternRobotics/G2VLM-2B-MoT/resolve/main/assets/method.png" width="100%"></p>

License

G2VLM is licensed under the Apache 2.0 license.

✍️ Citation

@article{hu2025g2vlmgeometrygroundedvision,
      title={G$^2$VLM: Geometry Grounded Vision Language Model with Unified 3D Reconstruction and Spatial Reasoning}, 
      author={Wenbo Hu and Jingli Lin and Yilin Long and Yunlong Ran and Lihan Jiang and Yifan Wang and Chenming Zhu and Runsen Xu and Tai Wang and Jiangmiao Pang},
      year={2025},
      eprint={2511.21688},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.21688}, 
}