---
license: mit
task_categories:
- object-detection
tags:
- ui
- user-interface
- web-design
- gui
pretty_name: Web UI Dataset
size_categories:
- n<1K
language:
- ru
- en
---

# Web UI Dataset

## Dataset Description

This dataset consists of real-world screenshots of web user interfaces, annotated for object detection. It is designed to be used in pair with the [Synthetic UI Dataset](https://huggingface.co/datasets/egnch/synthetic-ui) for fine-tuning and model validation.

- **Total Images:** 250
- **Task:** Object Detection
- **Source:** Annotation of websites screenshots

## Dataset Structure

### Classes
The dataset annotates **6 distinct classes** of UI elements:

1.  **button**: Buttons of various styles.
2.  **input**: Input fields, text areas, and dropdown lists.
3.  **checkbox**: Selection elements including checkboxes, radio buttons, and switches/toggles.
4.  **image**: Visual content including standard images, user avatars, and banners.
5.  **header**: Visual headers and highlights.
6.  **text**: Text content blocks, paragraphs, labels, and descriptions.

### Data Splits
The dataset is split into training and validation sets with a **90/10 ratio**:

| Split | Percentage | Number of Images |
| :--- | :--- | :--- |
| **Train** | 80% | 185 |
| **Validation** | 20% | 45 |

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("egnch/real-ui")

# Example: View the first training example
print(dataset["train"][0])
```

## Additional Information

### Annotation Distribution

| Class Label | Description | Count | Distribution (Approx.) | 
 | ----- | ----- | ----- | ----- |
| **text** | Text content blocks, paragraphs, labels | 2938 | ~50% | 
| **button** | Buttons of various styles | 955 | ~15% | 
| **image** | Visual content, avatars, banners | 508 | ~10% | 
| **header** | Visual headers and highlights | 446 | ~10% | 
| **input** | Input fields, text areas, dropdowns | 431 | ~10% | 
| **checkbox** | Checkboxes, radio buttons, switches | 263 | ~5% | 

### Licensing

This dataset is licensed under the MIT License.