UGround-V1-Data

## Updates

- **[May 1, 2025]** [**Bounding Box Data**](https://huggingface.co/datasets/osunlp/UGround-V1-Data-Box): We have added bounding box version of Web-Hybrid. For everyone's convenience, no conversation template is applied to this version of data. All the coordinates (x1, y1, x2, y2) are as always normalized to [0,999].

### Notes for Requests

If you have applied for access to this dataset but have not received approval, please contact us via email (Boyu Gou) with your name, institution, and research purpose.

Typically, requests will be approved within one day.

### Notes for Data

This repo contains the two datasets mentioned in our paper: Web-Hybrid and Web-Direct. The former is the primary source of performance gains.

For clarity, here is how we saved the data:
```
with open(item["image"], "rb") as img_file:
img_bytes = img_file.read()
record = {
"width": item["width"],
"height": item["height"],
"conversations": json.dumps(item["conversations"], ensure_ascii=False),
"image": img_bytes
}
```

The coordinates have been processed into Qwen2-VL's format, i,e, [0,999].

## 更新日志
- **[2025年5月1日]** [**边界框数据（Bounding Box Data）**](https://huggingface.co/datasets/osunlp/UGround-V1-Data-Box)：我们新增了Web-Hybrid的边界框版本。为便于所有使用者，该版本数据未应用任何对话模板。所有坐标(x1, y1, x2, y2)均统一归一化至[0,999]区间。

### 申请须知
若您已提交该数据集的访问权限申请但未获得审批，请通过邮件联系郭博宇（Boyu Gou），并提供您的姓名、所属机构及研究目的。
通常情况下，申请将在1个工作日内完成审批。

### 数据集说明
本仓库包含我们论文中提及的两个数据集：Web-Hybrid与Web-Direct。其中前者是模型性能提升的核心数据源。
为便于理解，我们的数据存储格式如下：

with open(item["image"], "rb") as img_file:
img_bytes = img_file.read()
record = {
"width": item["width"],
"height": item["height"],
"conversations": json.dumps(item["conversations"], ensure_ascii=False),
"image": img_bytes
}

所有坐标已处理为适配Qwen2-VL的格式，即归一化至[0,999]区间。

搜集汇总

背景与挑战

背景概述

UGround-V1-Data是由Web-Hybrid和Web-Direct组成的多模态数据集，特别包含经归一化处理的视觉定位数据。该数据集采用特定存储格式记录图像、尺寸及对话信息，其中Web-Hybrid子集对模型性能提升具有主要贡献。