easyr1-10k-hard-qwen7b-easy-gta1-4MP-dedup-pc-e-by-image-path

# easyr1-10k-hard-qwen7b-easy-gta1-4MP-dedup-pc-e-by-image-path

This dataset was generated using the EasyR1 grounding dataset pipeline.

## Generation Details

- **Generated on**: 2025-08-24 17:16:50 UTC
- **Script**: `push_easyr1_to_hf.py`
- **Data directory**: `/lustre/fsw/portfolios/nvr/users/aawadalla/LLaMA-Factory/data`

## Parameters Used

- **Maximum samples**: 10000
- **Image resize (max megapixels)**: 4.0 MP
- **Minimum native image resolution**: 0.0 MP
- **Prompt format**: `gta1_with_resolution`
- **Output format**: `coordinates`
- **Random seed**: 42
- **Resampling enabled**: False
- **Icon upsampling ratio**: Disabled (random sampling)
- **pc-agent-e deduplication**: True

## Dataset Groups

The following JSON/JSONL files were used to create this dataset:

### Dataset Group 1
Files (intersection of kept samples across all files):
- grounding-data-filters/pixmo-points-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl
- grounding-data-filters/pixmo-points-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl

### Dataset Group 2
Files (intersection of kept samples across all files):
- grounding-data-filters/autogui-grounding-only-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl
- grounding-data-filters/autogui-grounding-only-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl

### Dataset Group 3
Files (intersection of kept samples across all files):
- grounding-data-filters/seeclick-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl
- grounding-data-filters/seeclick-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl

### Dataset Group 4
Files (intersection of kept samples across all files):
- grounding-data-filters/pc-e-grounding-only-claude-instructions-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl
- grounding-data-filters/pc-e-grounding-only-claude-instructions-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl

### Dataset Group 5
Files (intersection of kept samples across all files):
- grounding-data-filters/omniact-grounding-only-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl
- grounding-data-filters/omniact-grounding-only-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl

### Dataset Group 6
Files (intersection of kept samples across all files):
- grounding-data-filters/showui-desktop-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl
- grounding-data-filters/showui-desktop-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl

### Dataset Group 7
Files (intersection of kept samples across all files):
- grounding-data-filters/showui-web-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl
- grounding-data-filters/showui-web-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl

### Dataset Group 8
Files (intersection of kept samples across all files):
- grounding-data-filters/uground-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl
- grounding-data-filters/uground-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl

### Dataset Group 9
Files (intersection of kept samples across all files):
- grounding-data-filters/waveui-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl
- grounding-data-filters/waveui-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl

## Dataset Statistics

- **Total training samples**: 10000
- **Image dimensions**: Variable
- **Columns**: image_path, prompt, normalized_bbox, images, easyr1_prompt, bbox, messages, image_width, image_height

## System Prompt

The following system prompt is used for this dataset:

```
You are an expert UI element locator. Given a GUI image and a user's element description, provide the coordinates of the specified element as a single (x,y) point. The image resolution is height 2048 and width 2048. For elements with area, return the center point.

Output the coordinate pair exactly:
(x,y)
```

## Sample Entry

- **User prompt**: <image>
Tap on badges, in the middle of the page
- **Assistant response**: (666,629)
- **Bounding box**: [208, 579, 1125, 680]
- **Image path**: uground-images/uground_034465.jpg

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP-dedup-pc-e-by-image-path")

# Access the training data
train_data = dataset['train']

# Example: Get the first sample
sample = train_data[0]
images = sample['images']
messages = sample['messages']
bbox = sample['bbox']
```

## Prompt Formats

### gta1_with_resolution

GTA1 format with image resolution included in the system prompt. Outputs coordinates in (x,y) format.

## License

Please refer to the original dataset licenses for usage restrictions.

# easyr1-10k-hard-qwen7b-easy-gta1-4MP-dedup-pc-e-by-image-path

本数据集基于EasyR1视觉锚定数据集流水线生成。

## 生成详情

- **生成时间**：2025-08-24 17:16:50 UTC
- **生成脚本**：`push_easyr1_to_hf.py`
- **数据目录**：`/lustre/fsw/portfolios/nvr/users/aawadalla/LLaMA-Factory/data`

## 所用参数

- **最大样本数**：10000
- **图像最大缩放分辨率**：4.0 百万像素（MP）
- **最小原生图像分辨率**：0.0 MP
- **提示格式**：`gta1_with_resolution`
- **输出格式**：坐标（coordinates）
- **随机种子**：42
- **重采样功能**：已禁用
- **图标上采样比例**：已禁用（采用随机采样）
- **pc-agent-e 去重**：已启用

## 数据集组

以下JSON/JSONL文件用于构建本数据集：

### 数据集组1
文件（所有文件中保留样本的交集）：
- `grounding-data-filters/pixmo-points-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl`
- `grounding-data-filters/pixmo-points-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl`

### 数据集组2
文件（所有文件中保留样本的交集）：
- `grounding-data-filters/autogui-grounding-only-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl`
- `grounding-data-filters/autogui-grounding-only-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl`

### 数据集组3
文件（所有文件中保留样本的交集）：
- `grounding-data-filters/seeclick-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl`
- `grounding-data-filters/seeclick-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl`

### 数据集组4
文件（所有文件中保留样本的交集）：
- `grounding-data-filters/pc-e-grounding-only-claude-instructions-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl`
- `grounding-data-filters/pc-e-grounding-only-claude-instructions-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl`

### 数据集组5
文件（所有文件中保留样本的交集）：
- `grounding-data-filters/omniact-grounding-only-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl`
- `grounding-data-filters/omniact-grounding-only-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl`

### 数据集组6
文件（所有文件中保留样本的交集）：
- `grounding-data-filters/showui-desktop-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl`
- `grounding-data-filters/showui-desktop-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl`

### 数据集组7
文件（所有文件中保留样本的交集）：
- `grounding-data-filters/showui-web-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl`
- `grounding-data-filters/showui-web-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl`

### 数据集组8
文件（所有文件中保留样本的交集）：
- `grounding-data-filters/uground-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl`
- `grounding-data-filters/uground-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl`

### 数据集组9
文件（所有文件中保留样本的交集）：
- `grounding-data-filters/waveui-qwen_tool_call-not_grounded-Qwen_Qwen2.5-VL-7B-Instruct-qwen_tool_call.jsonl`
- `grounding-data-filters/waveui-gta1-correctly_grounded-HelloKKMe_GTA1-7B-gta1.jsonl`

## 数据集统计信息

- **总训练样本数**：10000
- **图像尺寸**：可变
- **数据列**：image_path（图像路径）、prompt（提示词）、normalized_bbox（归一化边界框，normalized bounding box）、images（图像数据）、easyr1_prompt、bbox（边界框，bounding box）、messages（消息序列）、image_width（图像宽度）、image_height（图像高度）

## 系统提示词

你是一名专业的UI元素定位器（UI element locator）。给定一张GUI（图形用户界面，Graphical User Interface）图像与用户的元素描述，请以单个(x,y)坐标点的形式给出指定元素的位置。图像分辨率为高度2048、宽度2048。对于存在面积的元素，请返回其中心点坐标。

请严格按照以下格式输出坐标对：
(x,y)

## 样本示例

- **用户提示词**：<image> 点击页面中间的徽章
- **助手回复**：(666,629)
- **边界框**：[208, 579, 1125, 680]
- **图像路径**：uground-images/uground_034465.jpg

## 使用方法

python
from datasets import load_dataset

# 加载本数据集
dataset = load_dataset("mlfoundations-cua-dev/easyr1-10k-hard-qwen7b-easy-gta1-4MP-dedup-pc-e-by-image-path")

# 访问训练数据
train_data = dataset['train']

# 示例：获取第一个样本
sample = train_data[0]
images = sample['images']
messages = sample['messages']
bbox = sample['bbox']

## 提示格式

### gta1_with_resolution
包含图像分辨率的GTA1提示格式，输出为(x,y)格式的坐标。

## 许可证

请参考原始数据集的使用限制条款。