app.py

import gradio as gr
import spaces
from paddleocr import PaddleOCR
import fitz  # PyMuPDF
from PIL import Image
import numpy as np
import os

# --- 配置 ---
OUTPUT_DIR = "output_results" 
os.makedirs(OUTPUT_DIR, exist_ok=True)

# --- 模型加载器 ---
def load_gpu_model():
    print("正在加载经过版本锁定的PaddleOCR GPU模型...")
    # 使用与锁定版本兼容的参数：use_gpu=True是必需的
    ocr_model = PaddleOCR(use_angle_cls=True, lang='ch', use_gpu=True)
    print("GPU模型加载成功。")
    return ocr_model

# --- Gradio调用的核心处理函数 ---
@spaces.GPU
def process_pdf_max_speed(pdf_file, progress=gr.Progress(track_tqdm=True)):
    """
    使用GPU和批处理来极速处理PDF，并实时更新进度条。
    """
    if pdf_file is None:
        return "请先上传一个PDF文件。", None

    try:
        # 在GPU会话中加载模型
        ocr = load_gpu_model()
        
        doc = fitz.open(pdf_file.name)
        total_pages = len(doc)
        batch_size = 4  # 批处理大小
        full_text_result = []

        for i in progress.tqdm(range(0, total_pages, batch_size), desc="🚀 批处理中..."):
            
            batch_images = []
            for page_num in range(i, min(i + batch_size, total_pages)):
                page = doc.load_page(page_num)
                pix = page.get_pixmap(dpi=200)
                img = Image.frombytes("RGB", [pix.width, pix.height], pix.samples)
                batch_images.append(np.array(img))

            if batch_images:
                # 使用与锁定版本兼容的调用方式：需要 cls=True
                results = ocr.ocr(batch_images, cls=True)
                
                for page_index, page_result in enumerate(results):
                    page_texts = []
                    current_page_num = i + page_index + 1
                    if page_result:
                        for line in page_result:
                            page_texts.append(line[1][0])
                    
                    full_text_result.append(f"--- Page {current_page_num} ---\n" + "\n".join(page_texts))

        doc.close()
        
        final_text = "\n\n".join(full_text_result)
        output_filename = os.path.join(OUTPUT_DIR, f"{os.path.splitext(os.path.basename(pdf_file.name))[0]}_result.txt")
        with open(output_filename, 'w', encoding='utf-8') as f:
            f.write(final_text)

        print(f"处理完成！结果已保存到 {output_filename}")
        return final_text, output_filename

    except Exception as e:
        error_message = f"处理过程中发生错误: {str(e)}"
        print(error_message)
        return error_message, None

# --- 构建Gradio界面 ---
with gr.Blocks(title="极速PDF识别器", theme=gr.themes.Soft()) as demo:
    gr.Markdown(
        """
        # 🔥 极速PDF识别器 (GPU加速稳定版) 🔥
        **速度拉满！实时进度显示，处理期间请勿关闭页面。**
        """
    )
    
    with gr.Row():
        pdf_input = gr.File(label="📄 上传PDF文件", file_types=[".pdf"])
    
    submit_btn = gr.Button("⚡️ 开始极速处理", variant="primary")
    
    result_display = gr.Textbox(label="识别结果", lines=20, show_copy_button=True)
    download_link = gr.File(label="📥 点击此处下载结果文件", interactive=False)

    submit_btn.click(
        fn=process_pdf_max_speed,
        inputs=[pdf_input],
        outputs=[result_display, download_link]
    )

demo.queue().launch()