README.md

---
title: DeepSeek-OCR Studio
emoji: 🔍
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: mit
---

# 🔍 DeepSeek-OCR Studio

Advanced OCR system based on [DeepSeek-OCR](https://github.com/deepseek-ai/DeepSeek-OCR), providing powerful document parsing capabilities.

## ✨ Features

- **Multi-language OCR**: Support for Chinese, English, and many other languages
- **Table Extraction**: Intelligent table recognition and markdown conversion
- **Chart Analysis**: Extract data from charts and graphs
- **Professional Drawings**: Semantic recognition of CAD drawings, flowcharts, etc.
- **Layout Analysis**: Preserve document structure and formatting
- **PDF Support**: Process PDF documents page by page

## 🚀 Quick Start

1. **Upload an Image or PDF**: Click to upload your document
2. **Optional Prompt**: Customize the OCR task (e.g., "Extract tables", "Analyze chart")
3. **Extract Text**: Click the button and wait for results

## 📝 Prompt Examples

### Basic OCR
```
Free OCR.
```

### Table Extraction
```
Extract all tables and convert to markdown format.
```

### Chart Analysis
```
Analyze this chart and extract data in table format.
```

### Multi-language Documents
```
Extract all text in multiple languages.
```

### Technical Drawings
```
Analyze this CAD drawing and describe its components.
```

## ⚙️ Deployment Information

- **Platform**: Hugging Face Spaces with ZeroGPU
- **Model**: [deepseek-ai/DeepSeek-OCR](https://huggingface.co/deepseek-ai/DeepSeek-OCR)
- **Processing Time**: 30-120 seconds per image/page
- **PDF Limitation**: First 3 pages only (ZeroGPU constraint)

## 🔧 Local Deployment

For full functionality with unlimited pages and faster processing:

```bash
# Clone the repository
git clone https://github.com/fufankeji/DeepSeek-OCR-Web.git
cd DeepSeek-OCR-Web

# Install dependencies
bash install.sh

# Start services
bash start.sh
```

**Requirements**:
- Linux OS
- GPU with ≥7GB VRAM (16-24GB recommended)
- Python 3.10-3.12
- CUDA 11.8 or 12.1/12.2

## 📚 Documentation

- [Official DeepSeek-OCR](https://github.com/deepseek-ai/DeepSeek-OCR)
- [Web Interface Repository](https://github.com/fufankeji/DeepSeek-OCR-Web)
- [Model on Hugging Face](https://huggingface.co/deepseek-ai/DeepSeek-OCR)

## 🙏 Acknowledgments

- **DeepSeek AI**: For the amazing OCR model
- **Hugging Face**: For providing ZeroGPU infrastructure
- Original project: [DeepSeek-OCR-Web](https://github.com/fufankeji/DeepSeek-OCR-Web)

## 📄 License

MIT License

## 🐛 Known Limitations on Spaces

- ZeroGPU has time limits (120 seconds per request)
- PDF processing limited to first 3 pages
- First request takes longer (model loading)
- Large images may timeout

For production use, please deploy locally with dedicated GPU.