GSEval

# 🚀 GSEval - A Comprehensive Grounding Evaluation Benchmark

GSEval is a meticulously curated evaluation benchmark consisting of 3,800 images, designed to assess the performance of pixel-level and bounding box-level grounding models. It evaluates how well AI systems can understand and localize objects or regions in images based on natural language descriptions.

## 📊 Results
<div align="center">
<img src="./assets/gseval.png">
</div>

<div align="center">
<img src="./assets/gseval_box.png">
</div>

## 📊 Download GSEval
```
git lfs install
git clone https://huggingface.co/datasets/hustvl/GSEval
```

## 📚 Additional Resources
- **Paper:** [ArXiv](https://arxiv.org/abs/2503.10596)
- **GitHub Repository:** [GitHub - GroundingSuite](https://github.com/hustvl/GroundingSuite)

# 🚀 GSEval——全方位接地评估基准数据集

GSEval是一款精心遴选打造的评估基准数据集，包含3800张图像，旨在评测像素级（pixel-level）与边界框级（bounding box-level）接地模型（grounding model）的性能。该基准用于评估人工智能系统基于自然语言描述理解图像并定位目标或区域的能力。

## 📊 实验结果
<div align="center">
<img src="./assets/gseval.png">
</div>

<div align="center">
<img src="./assets/gseval_box.png">
</div>

## 📊 下载GSEval

git lfs install
git clone https://huggingface.co/datasets/hustvl/GSEval

## 📚 附加资源
- **论文:** [ArXiv预印本](https://arxiv.org/abs/2503.10596)
- **GitHub仓库:** [GroundingSuite](https://github.com/hustvl/GroundingSuite)