Home

PaddleOCR‑VL Local UI — Wiki Home

Welcome! This wiki is the living documentation for the project.

Disclaimer
This project was built with the help of AI, and I am not a professional software engineer.
Please treat it as a personal/experimental project. Use at your own risk.

---

1) Project Status

Current state (MVP complete):

• Local web UI (localhost)
• Drag & drop uploads (PDF/images)
• Job queue with logs + history
• Outputs + preview + ZIP download
• Docker‑based inference using PaddleOCR‑VL (v1)

Known limitations:

• No DOCX or searchable PDF export yet
• Single‑worker queue (1 job at a time)
• No authentication (local only)

---

2) Quick Start

Requirements

• Docker Desktop (running)
• Python 3.10+
• PaddleOCR‑VL Docker image:

  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu
  

Setup (Windows)

cd C:\Users\egor0\Downloads\WebforAI\publish
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r server\requirements.txt

Run

uvicorn server.app:app --host 127.0.0.1 --port 8000

Open: http://127.0.0.1:8000

---

3) How It Works (High‑Level)

Browser → FastAPI → Docker → PaddleOCR‑VL → results → Browser

Data flow

1. Upload file in UI
2. Server stores file under data/jobs/<job_id>/input
3. Worker runs Docker paddleocr doc_parser
4. Outputs saved to data/jobs/<job_id>/output
5. UI shows outputs + preview + logs

---

4) Folder Structure

publish/
  server/        # API + queue + docker execution
  web/           # HTML/CSS/JS UI
  data/
    jobs/        # outputs by job

---

5) Features

Upload

• PDF, PNG, JPG, TIFF, WEBP
• Multiple files
• Remove files before Start

Outputs

• JSON (*_res.json)
• Markdown (*.md)
• Plain text (*_text.txt)
• Layout image (*_layout_det_res.jpg)
• ZIP download

Logs

• Live log streaming
• Colored highlights

---

6) Troubleshooting

Docker says “not running”

docker info

Start Docker Desktop and retry.

Image missing

• Use “Pull image” in UI
• Or run:

docker pull ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu

Model download is slow First run downloads model weights into:

C:\Users\egor0\.paddleocr-vl-cache

Next runs are faster.

---

7) Roadmap (Suggested)

Near term

• DOCX export
• Searchable PDF export
• Cancel running job
• Multi‑worker queue

Later

• Remote access + auth
• Cloud deployment
• User accounts / session history

---

8) Contributing

PRs are welcome!
If you improve UI/UX, add features, or fix bugs, feel free to open a PR.

---

9) Security

See SECURITY.md (or SECURITY_TEMPLATE.md) for reporting vulnerabilities.