A real-time Electron-based desktop GUI for DeepSeek-OCR

Unaffiliated with DeepSeek

• Drag-and-drop image upload
• Real-time OCR processing

• Click regions to copy
• Export results as ZIP with markdown images
• GPU acceleration (CUDA)

• Windows 10/11, other OS are experimental
• Node.js 18+ (download)
• Python 3.12+ (download)
• NVIDIA GPU with CUDA

conda create -yn ocr python=3.12 conda activate ocr pip install -r requirements.txt #重要： $env:ELECTRON_MIRROR="https://npmmirror.com/mirrors/electron/" npm install npm install --registry=https://registry.npmjs.org/

https://registry.npmmirror.com

You are using a model of type deepseek_v1_v2 to instantiate a model of type DeepseekOCR. This is not supported for all configurations of models and can yield errors.

你正在使用 deepseek_v1_v2 类型的模型来实例化 DeepseekOCR 类型的模型。这在某些配置下不被支持，可能会产生错误。

原因 这是 Hugging Face Transformers 库的兼容性警告：

DeepSeek-OCR 模型的配置文件中定义的模型类型是 deepseek_v1_v2 但实际加载时使用的是 DeepseekOCR 类 Transformers 检测到类型不完全匹配，发出警告 这是官网 https://deepseekocr.com/zh-CN#app

1. Extract the ZIP file
2. Run start-client.bat
   • First run will automatically install dependencies.
   • Subsequent runs will start quicker.
3. Load Model - Click the "Load Model" button in the app, this will download or load the model.
   • If this is the first run, this might take some time.
4. Drop an image or click the drop zone to select one.
5. Run OCR - Click "Run OCR" to process.

Note: if you have issues processing images but the model loads properly, please close and re-open the app and try with the default resolution for "base" and "size". This is a known issue, if you can help to fix it I would appreciate it!

Note: Linux and macOS have not been tested yet. Use start-client.sh instead of start-client.bat.

PRs welcome! If you test on Linux/macOS and encounter issues, please open a pull request with fixes.

• Model HuggingFace
• Model Blog Post
• Model GitHub

• Code cleanup needed (quickly put together)
• TypeScript
• Updater from GitHub releases
• PDF support
• Batch processing
• CPU support?
• Web version (so you can run the server on a different machine)
• Better progress bar algo
• ???

MIT