DeepSeek-OCR Web Playground 🚀🖼️

English | 中文

FastAPI + Vite/Vue wrapper around the DeepSeek-OCR model for quick local testing. 🎯

Showcase:

Showcase: Mathematical Formula Recognition

Showcase: Table Data Processing

Features ✨

• 📄 PDF & 🖼️ image upload, with automatic PDF page splitting.
• ⏳ Progress visualization during uploads/inference so you know it’s working.
• 🗺️ Bounding-box overlay for layout/annotation visualization.
• 📑 Annotation detail view for extracted labels/regions.
• 📜 History management: View and manage past OCR results.
• 🧩 Modular UI: Specialized components for Upload, Prompt, Mode, and Actions.
• 📊 Enhanced Visualization: Visual preview, layout details, and page text inspection.
• 📝 Logs & Workflow: Real-time logs and workflow step tracking.

Screenshots:

• Overlay view:
• Annotation details:

Tested environment 🧪

• Conda env: ds-ocr
• Python 3.10.19 (conda-forge)
• PyTorch 2.6.0+cu118 (CUDA)
• GPU: NVIDIA GeForce RTX 3090

Quick start (conda) ⚡

# 1) create / activate
conda create -n ds-ocr python=3.10 -y
conda activate ds-ocr

# 2) backend deps (install torch CUDA build first if needed)
cd web_project/backend
pip install --upgrade pip
# offline / matching CUDA build:
# pip install --no-index --find-links ../../wheels torch torchvision
pip install -r requirements.txt

# 3) frontend deps
cd ../frontend
npm install

# 4) run dev (from repo root)
cd ../../
./start.sh

./start.sh starts FastAPI on localhost:8000 and Vite dev server on localhost:5173. 🌐

Run separately 🧭

# backend
cd web_project/backend
uvicorn app:app --host 0.0.0.0 --port 8000

# frontend
cd web_project/frontend
npm run dev

Key env vars 🛠️

• OCR_BACKEND_PORT: backend port (default 8000)
• DEEPSEEK_OCR_MODEL_PATH: model weights path override (default ocr_project/model)
• DEEPSEEK_ATTN_IMPL: attention impl, e.g. flash_attention_2
• DEEPSEEK_ALLOWED_ORIGINS: CORS allowlist, comma-separated
• DEEPSEEK_MAX_IMAGE_MB: upload size limit in MB (default 15)