ocr_verify/ocr_vlm_verify/simple_ocr_test.py

#!/usr/bin/env python3
"""
简化版OCR验证脚本 - 获取原始分析内容
"""

import os
import base64
import json
from pathlib import Path
from openai import OpenAI
from dotenv import load_dotenv

# 加载环境变量
load_dotenv()

def simple_ocr_verification():
    """简化的OCR验证，返回原始分析文本"""
    
    # 获取配置
    api_key = os.getenv("YUSYS_MULTIMODAL_API_KEY")
    api_base = os.getenv("YUSYS_MULTIMODAL_API_BASE")
    model_id = os.getenv("YUSYS_MULTIMODAL_ID")
    model_name = model_id.replace("openai/", "") if model_id else ""
    
    # 文件路径
    image_path = "工大照片-1.jpg"
    ocr_json_path = "demo_54fa7ad0_page_1.json"
    
    # 读取图片
    with open(image_path, "rb") as f:
        image_data = base64.b64encode(f.read()).decode('utf-8')
    
    # 读取OCR结果
    with open(ocr_json_path, "r", encoding='utf-8') as f:
        ocr_results = json.load(f)
    
    # 构建OCR文本摘要
    ocr_summary = f"OCR识别了{len(ocr_results)}个项目:\\n"
    for i, item in enumerate(ocr_results[:5], 1):  # 只显示前5个
        bbox = item.get('bbox', [])
        text = item.get('text', '')[:50]  # 限制文本长度
        ocr_summary += f"{i}. 位置{bbox} - 文本: {text}\\n"
    if len(ocr_results) > 5:
        ocr_summary += f"... 还有{len(ocr_results) - 5}个项目\\n"
    
    # 简化的提示词
    prompt = f"""请分析这张图片，并与OCR识别结果进行对比。

{ocr_summary}

请详细描述：
1. 图片的内容是什么？
2. OCR识别的结果是否与图片内容匹配？
3. 你发现了哪些明显的错误或问题？
4. 图片与OCR结果是否匹配同一份文档？

请用中文详细回答，不需要JSON格式。"""
    
    # 构建消息
    messages = [
        {
            "role": "user",
            "content": [
                {"type": "text", "text": prompt},
                {
                    "type": "image_url",
                    "image_url": {
                        "url": f"data:image/jpeg;base64,{image_data}"
                    }
                }
            ]
        }
    ]
    
    # 调用API
    client = OpenAI(api_key=api_key, base_url=api_base)
    
    print("正在分析图片和OCR结果...")
    response = client.chat.completions.create(
        model=model_name,
        messages=messages,  # type: ignore
        temperature=0.3,
        max_tokens=2048
    )
    
    analysis = response.choices[0].message.content or "未获取到分析结果"
    
    print("\\n=== VLM详细分析结果 ===")
    print(analysis)
    
    # 保存原始分析
    with open("raw_analysis.txt", "w", encoding="utf-8") as f:
        f.write(f"图片: {image_path}\\n")
        f.write(f"OCR文件: {ocr_json_path}\\n")
        f.write(f"模型: {model_name}\\n")
        f.write(f"OCR结果摘要:\\n{ocr_summary}\\n\\n")
        f.write("=== VLM分析 ===\\n")
        f.write(analysis)
    
    print(f"\\n原始分析已保存到: raw_analysis.txt")

if __name__ == "__main__":
    simple_ocr_verification()