---
base_model:
- OpenGVLab/InternVL2-8B
license: apache-2.0
library_name: transformers
pipeline_tag: image-text-to-text
---

## SpiritSight Agent: Advanced GUI Agent with One Look

<p align="center">
    <a href="https://arxiv.org/abs/2503.03196">📄 Paper</a> •
    <a href="https://huggingface.co/SenseLLM/SpiritSight-Agent-8B">🤖 Models</a> •
    <a href="https://hzhiyuan.github.io/SpiritSight-Agent"> 🏠 Project Page</a> •
    <a href="https://huggingface.co/datasets/SenseLLM/GUI-Lasagne-L1">📚 Datasets</a>
</p>


## Introduction

SpiritSight-Agent is a vision-based, end-to-end GUI agent that excels in GUI navigation tasks across various GUI platforms. It leverages advanced Vision Language Models (VLMs) to achieve high accuracy, low latency, and compatibility for different GUI platforms.

## Paper Abstract

Graphical User Interface (GUI) agents show amazing abilities in assisting human-computer interaction, automating human user's navigation on digital devices. An ideal GUI agent is expected to achieve high accuracy, low latency, and compatibility for different GUI platforms. Recent vision-based approaches have shown promise by leveraging advanced Vision Language Models (VLMs). While they generally meet the requirements of compatibility and low latency, these vision-based GUI agents tend to have low accuracy due to their limitations in element grounding. To address this issue, we propose **SpiritSight**, a vision-based, end-to-end GUI agent that excels in GUI navigation tasks across various GUI platforms. First, we create a multi-level, large-scale, high-quality GUI dataset called **GUI-Lasagne** using scalable methods, empowering SpiritSight with robust GUI understanding and grounding capabilities. Second, we introduce the **Universal Block Parsing (UBP)** method to resolve the ambiguity problem in dynamic high-resolution of visual inputs, further enhancing SpiritSight's ability to ground GUI objects. Through these efforts, SpiritSight agent outperforms other advanced methods on diverse GUI benchmarks, demonstrating its superior capability and compatibility in GUI navigation tasks.


![](results.png)
![](results2.png)


## Models

We recommend fine-tuning the base model on custom data.

| Model | Checkpoint | Size | License|
|:-------|:------------|:------|:--------|
| SpiritSight-Agent-2B-base  | 🤗 [HF Link](https://huggingface.co/SenseLLM/SpiritSight-Agent-2B)  | 2B  | [InternVL](https://github.com/OpenGVLab/InternVL/blob/main/LICENSE) |
| SpiritSight-Agent-8B-base  | 🤗 [HF Link](https://huggingface.co/SenseLLM/SpiritSight-Agent-8B)  | 8B  | [InternVL](https://github.com/OpenGVLab/InternVL/blob/main/LICENSE) |
| SpiritSight-Agent-26B-base | 🤗 [HF Link](https://huggingface.co/SenseLLM/SpiritSight-Agent-26B) | 26B | [InternVL](https://github.com/OpenGVLab/InternVL/blob/main/LICENSE) |


## Datasets

Coming soon.


## Inference

```shell
conda create -n spiritsight-agent python=3.9

pip install -r requirements.txt
pip install flash-attn==2.3.6 --no-build-isolation

python infer_SSAgent-8B.py
```


## Citation

If you find this repo useful for your research, please kindly cite our paper:
```
@misc{huang2025spiritsightagentadvancedgui,
      title={SpiritSight Agent: Advanced GUI Agent with One Look}, 
      author={Zhiyuan Huang and Ziming Cheng and Junting Pan and Zhaohui Hou and Mingjie Zhan},
      year={2025},
      eprint={2503.03196},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.03196},
}
```


## Acknowledgments

We thank the following amazing projects that truly inspired us:

- [InternVL2](https://huggingface.co/OpenGVLab/InternVL2-8B)
- [SeeClick]( https://github.com/njucckevin/SeeClick)
- [Mind2Web](https://huggingface.co/datasets/osunlp/Multimodal-Mind2Web)
- [GUI-Odyssey](https://github.com/OpenGVLab/GUI-Odyssey)
- [AMEX](https://huggingface.co/datasets/Yuxiang007/AMEX)
- [AndroidControl](https://github.com/google-research/google-research/tree/master/android_control)
- [GUICourse](https://github.com/yiye3/GUICourse)