Image Recognition System v3.0

PP-OCRv6 ONNX + YOLO26n + ImageCache Smart OCR System. TCP Socket based architecture.

Architecture

                    HTTP 11719
  +------------------+------------------+
  |                                     |
  v                                     v
+---------------+     TCP 11720     +---------------+
|  API Service  | <---------------> |  SOE Server   |
|  (FastAPI)    |                   |  (TCP Socket) |
+---------------+                   +---------------+
  |                                     |
  | /v1/* (OpenAI)                     | OCR Engine
  | /* (Standard JSON)                  | YOLO Detector
  +------------------+                  | Image Cache
                     |                  | Device Profiler
                     v                  |
              +-------------+           |
              |   Client    |           |
              |  (Adapter)  | <---------+
              +-------------+

File List

File	Description
ocr_openai_api.py	API Service (HTTP), dual mode: v1 OpenAI + Standard JSON
smart_ocr_onnx.py	SOE TCP Server (TCP 11720), core OCR engine
adapter.py	TCP Client, connects to SOE via TCP
model_downloader.py	Multi-source auto downloader
image_cache.py	Image cache (SHA1 + pHash, SQLite)
yolo_detector.py	YOLO26n detector (ONNX/OpenVINO)
device_profiler.py	Device profiling & dynamic config
onnx_ocr_engine.py	ONNX OCR engine
ocr_utils.py	OCR utilities
ocr_compat_onnx.py	Compat entry (TCP client)
compare_images_gt730_v34.py	Deep feature comparison
convert_models.py	Model conversion
launcher.py	Launcher v3.0
config.json	Full config with TCP settings
requirements.txt	Dependencies

Quick Start

1. Install Dependencies

pip install -r requirements.txt
# Optional: pip install openvino
# Optional: pip install onnxruntime-gpu

2. Start SOE TCP Server

python smart_ocr_onnx.py --port 11720
# or
python launcher.py --soe

3. Start API Service (another terminal)

python ocr_openai_api.py --port 11719 --soe-port 11720
# or
python launcher.py --api

4. Test

# Standard API - OCR
curl -X POST http://localhost:11719/ocr \
  -H "Content-Type: application/json" \
  -d '{"image_path": "test.jpg"}'

# Standard API - YOLO
curl -X POST http://localhost:11719/yolo \
  -H "Content-Type: application/json" \
  -d '{"image_path": "test.jpg"}'

# OpenAI v1 API
curl -X POST http://localhost:11719/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ocr-ppocr-v4",
    "messages": [{"role": "user", "content": "recognize"}],
    "image_url": "https://example.com/image.jpg"
  }'

Standard API Endpoints

Method	Path	Description
GET	/	Service info
GET	/health	Health check
GET	/models	Model list
POST	/ocr	Full OCR recognition
POST	/yolo	YOLO detection only
GET	/cache/stats	Cache statistics
POST	/cache/clear	Clear cache
GET	/profile	Device profile
POST	/ping	Ping SOE

OpenAI v1 API Endpoints

Method	Path	Description
GET	/v1/models	Model list
POST	/v1/chat/completions	Chat completion (OCR)

TCP Protocol

SOE TCP Server listens on 127.0.0.1:11720.

Request format: JSON + newline

{"cmd": "ocr", "image_path": "test.jpg", "image_id": "test"}

Response format: JSON + newline

{"status": "ok", "image_id": "test", "content": [...], "yolo": [...]}

Commands:

• ocr - Recognize image
• cache_stats - Cache statistics
• cache_clear - Clear cache
• profile - Device profile
• ping - Heartbeat
• exit - Shutdown

Config

Key settings in config.json:

{
  "tcp": {
    "soe_host": "127.0.0.1",
    "soe_port": 11720,
    "api_host": "0.0.0.0",
    "api_port": 11719
  }
}

CLI Usage

# SOE Server
python smart_ocr_onnx.py --host 127.0.0.1 --port 11720

# API Service
python ocr_openai_api.py --host 0.0.0.0 --port 11719 --soe-host 127.0.0.1 --soe-port 11720

# Single image (auto starts SOE)
python smart_ocr_onnx.py image.jpg

# Download only
python smart_ocr_onnx.py --download

Detection Classes

YOLO26n supports: text, table, formula, handwriting, stamp, qr_code, barcode.