DeepSeek-OCR Web Playground
A FastAPI and Vue local web UI wrapper around the DeepSeek-OCR model, letting a user upload a PDF or image, watch inference progress, and inspect the extracted text with bounding-box overlays, including math formulas and tables.
This project is an out-of-the-box local web UI for the DeepSeek-OCR model, wrapping it with a FastAPI backend and a Vite and Vue frontend for quick local testing. It supports PDF and image uploads with automatic PDF page splitting, and shows upload and inference progress so the user can see the job is running.
Results are shown with a bounding-box overlay for layout and annotation, a detail view for extracted labels and regions, and history management to review past OCR runs; the model has been shown handling mathematical formula recognition and table data extraction. The UI is split into modular components for upload, prompt, mode and actions, with additional visualization for layout details and page-text inspection, plus real-time logs and workflow step tracking.
It is tested with a Conda environment running Python 3.10 and PyTorch 2.6 with CUDA on an NVIDIA GeForce RTX 3090, starting the backend on port 8000 and a Vite dev server on port 5173, with environment variables for the backend port, model path, attention implementation, CORS allowlist and upload size limit.