""" 批量处理图片并生成符合评测要求的预测结果 根据 OmniDocBench 评测要求： - 输入：OpenDataLab___OmniDocBench/images 下的所有 .jpg 图片，以及PDF文件 - 输出：每个图片对应的 .md、.json 和带标注的 layout 图片文件 - 输出目录：用于后续的 end2end 评测 """ import os import sys import json import tempfile import uuid import shutil import time import traceback import warnings from pathlib import Path from typing import List, Dict, Any from PIL import Image from tqdm import tqdm import argparse # 导入 dots.ocr 相关模块 from dots_ocr.parser import DotsOCRParser from dots_ocr.utils import dict_promptmode_to_prompt from dots_ocr.utils.consts import MIN_PIXELS, MAX_PIXELS from dots_ocr.utils.doc_utils import load_images_from_pdf # 导入工具函数 from utils import ( get_image_files_from_dir, get_image_files_from_list, get_image_files_from_csv, collect_pid_files, normalize_markdown_table, normalize_json_table ) def convert_pdf_to_images(pdf_file: str, output_dir: str | None = None, dpi: int = 200) -> List[str]: """ 将PDF转换为图像文件 Args: pdf_file: PDF文件路径 output_dir: 输出目录 dpi: 图像分辨率 Returns: 生成的图像文件路径列表 """ pdf_path = Path(pdf_file) if not pdf_path.exists() or pdf_path.suffix.lower() != '.pdf': print(f"❌ Invalid PDF file: {pdf_path}") return [] # 如果没有指定输出目录，使用PDF同名目录 if output_dir is None: output_path = pdf_path.parent / f"{pdf_path.stem}" else: output_path = Path(output_dir) / f"{pdf_path.stem}" output_path = output_path.resolve() output_path.mkdir(parents=True, exist_ok=True) try: # 使用utils中的函数加载PDF图像 images = load_images_from_pdf(str(pdf_path), dpi=dpi) image_paths = [] for i, image in enumerate(images): # 生成图像文件名 image_filename = f"{pdf_path.stem}_page_{i+1:03d}.png" image_path = output_path / image_filename # 保存图像 image.save(str(image_path)) image_paths.append(str(image_path)) print(f"✅ Converted {len(images)} pages from {pdf_path.name} to images") return image_paths except Exception as e: print(f"❌ Error converting PDF {pdf_path}: {e}") traceback.print_exc() return [] def get_input_files(args) -> List[str]: """ 获取输入文件列表，统一处理PDF和图像文件 Args: args: 命令行参数 Returns: 处理后的图像文件路径列表 """ input_files = [] # 获取原始输入文件 if args.input_csv: raw_files = get_image_files_from_csv(args.input_csv, "fail") elif args.input_file_list: raw_files = get_image_files_from_list(args.input_file_list) elif args.input_file: raw_files = [Path(args.input_file).resolve()] else: input_dir = Path(args.input_dir).resolve() if not input_dir.exists(): print(f"❌ Input directory does not exist: {input_dir}") return [] # 获取所有支持的文件（图像和PDF） image_extensions = ['.jpg', '.jpeg', '.png', '.bmp', '.tiff', '.tif'] pdf_extensions = ['.pdf'] raw_files = [] for ext in image_extensions + pdf_extensions: raw_files.extend(list(input_dir.glob(f"*{ext}"))) raw_files.extend(list(input_dir.glob(f"*{ext.upper()}"))) raw_files = [str(f) for f in raw_files] # 分别处理PDF和图像文件 pdf_count = 0 image_count = 0 for file_path in raw_files: file_path = Path(file_path) if file_path.suffix.lower() == '.pdf': # 转换PDF为图像 print(f"📄 Processing PDF: {file_path.name}") pdf_images = convert_pdf_to_images( str(file_path), args.output_dir, dpi=args.dpi ) input_files.extend(pdf_images) pdf_count += 1 else: # 直接添加图像文件 if file_path.exists(): input_files.append(str(file_path)) image_count += 1 print(f"📊 Input summary:") print(f" PDF files processed: {pdf_count}") print(f" Image files found: {image_count}") print(f" Total image files to process: {len(input_files)}") return input_files class DotsOCRProcessor: """DotsOCR 处理器""" def __init__(self, ip: str = "127.0.0.1", port: int = 8101, model_name: str = "DotsOCR", prompt_mode: str = "prompt_layout_all_en", dpi: int = 200, min_pixels: int = MIN_PIXELS, max_pixels: int = MAX_PIXELS, normalize_numbers: bool = False): """ 初始化处理器 Args: ip: vLLM 服务器 IP port: vLLM 服务器端口 model_name: 模型名称 prompt_mode: 提示模式 dpi: PDF 处理 DPI min_pixels: 最小像素数 max_pixels: 最大像素数 """ self.ip = ip self.port = port self.model_name = model_name self.prompt_mode = prompt_mode self.dpi = dpi self.min_pixels = min_pixels self.max_pixels = max_pixels self.normalize_numbers = normalize_numbers # 初始化解析器 self.parser = DotsOCRParser( ip=ip, port=port, dpi=dpi, min_pixels=min_pixels, max_pixels=max_pixels, model_name=model_name ) print(f"DotsOCR Parser 初始化完成:") print(f" - 服务器: {ip}:{port}") print(f" - 模型: {model_name}") print(f" - 提示模式: {prompt_mode}") print(f" - 像素范围: {min_pixels} - {max_pixels}") def create_temp_session_dir(self) -> tuple: """创建临时会话目录""" session_id = uuid.uuid4().hex[:8] temp_dir = os.path.join(tempfile.gettempdir(), f"omnidocbench_batch_{session_id}") os.makedirs(temp_dir, exist_ok=True) return temp_dir, session_id def save_results_to_output_dir(self, result: Dict, image_name: str, output_dir: str) -> Dict[str, str]: """ 将处理结果保存到输出目录 Args: result: 解析结果 image_name: 图片文件名（不含扩展名） output_dir: 输出目录 Returns: dict: 保存的文件路径 """ saved_files = {} try: # 1. 保存 Markdown 文件（OmniDocBench 评测必需） output_md_path = os.path.join(output_dir, f"{image_name}.md") md_content = "" # 优先使用无页眉页脚的版本（符合 OmniDocBench 评测要求） if 'md_content_nohf_path' in result and os.path.exists(result['md_content_nohf_path']): with open(result['md_content_nohf_path'], 'r', encoding='utf-8') as f: md_content = f.read() elif 'md_content_path' in result and os.path.exists(result['md_content_path']): with open(result['md_content_path'], 'r', encoding='utf-8') as f: md_content = f.read() else: md_content = "# 解析失败\n\n未能提取到有效的文档内容。" # 如果启用数字标准化，处理 Markdown 内容 original_text = md_content if self.normalize_numbers: # generated_text = normalize_financial_numbers(generated_text) # 只对Markdown表格进行数字标准化 generated_text = normalize_markdown_table(md_content) # 统计标准化的变化 changes_count = len([1 for o, n in zip(original_text, generated_text) if o != n]) if changes_count > 0: saved_files['md_normalized'] = f"✅ 已标准化 {changes_count} 个字符（全角→半角）" else: saved_files['md_normalized'] = "ℹ️ 无需标准化（已是标准格式）" with open(output_md_path, 'w', encoding='utf-8') as f: f.write(generated_text) saved_files['md'] = output_md_path # 如果启用了标准化，也保存原始版本用于对比 if self.normalize_numbers and original_text != generated_text: original_markdown_path = Path(output_dir) / f"{Path(image_name).stem}_original.md" with open(original_markdown_path, 'w', encoding='utf-8') as f: f.write(original_text) # 2. 保存 JSON 文件 output_json_path = os.path.join(output_dir, f"{image_name}.json") json_data = {} if 'layout_info_path' in result and os.path.exists(result['layout_info_path']): with open(result['layout_info_path'], 'r', encoding='utf-8') as f: json_content = f.read() else: json_content = f'{{"error": "未能提取到有效的布局信息"}}' # 对json中的表格内容进行数字标准化， original_json_text = json_content if self.normalize_numbers: json_content = normalize_json_table(json_content) # 统计标准化的变化 changes_count = len([1 for o, n in zip(original_json_text, json_content) if o != n]) if changes_count > 0: saved_files['json_normalized'] = f"✅ 已标准化 {changes_count} 个字符（全角→半角）" else: saved_files['json_normalized'] = "ℹ️ 无需标准化（已是标准格式）" with open(output_json_path, 'w', encoding='utf-8') as f: f.write(json_content) saved_files['json'] = output_json_path # 如果启用了标准化，也保存原始版本用于对比 if self.normalize_numbers and original_json_text != json_content: original_json_path = Path(output_dir) / f"{Path(image_name).stem}_original.json" with open(original_json_path, 'w', encoding='utf-8') as f: f.write(original_json_text) # 3. 保存带标注的布局图片 output_layout_image_path = os.path.join(output_dir, f"{image_name}_layout.jpg") if 'layout_image_path' in result and os.path.exists(result['layout_image_path']): # 直接复制布局图片 shutil.copy2(result['layout_image_path'], output_layout_image_path) saved_files['layout_image'] = output_layout_image_path else: # 如果没有布局图片，使用原始图片作为占位符 try: original_image = Image.open(result.get('original_image_path', '')) original_image.save(output_layout_image_path, 'JPEG', quality=95) saved_files['layout_image'] = output_layout_image_path except Exception as e: saved_files['layout_image'] = None except Exception as e: print(f"Error saving results for {image_name}: {e}") return saved_files def process_single_image(self, image_path: str, output_dir: str) -> Dict[str, Any]: """ 处理单张图片 Args: image_path: 图片路径 output_dir: 输出目录 Returns: dict: 处理结果 """ start_time = time.time() image_name = Path(image_path).stem result_info = { "image_path": image_path, "processing_time": 0, "success": False, "device": f"{self.ip}:{self.port}", "error": None, "output_files": {}, "is_pdf_page": "_page_" in Path(image_path).name # 标记是否为PDF页面 } try: # 检查输出文件是否已存在 output_md_path = os.path.join(output_dir, f"{image_name}.md") output_json_path = os.path.join(output_dir, f"{image_name}.json") output_layout_path = os.path.join(output_dir, f"{image_name}_layout.jpg") if all(os.path.exists(p) for p in [output_md_path, output_json_path, output_layout_path]): result_info.update({ "success": True, "processing_time": 0, "output_files": { "md": output_md_path, "json": output_json_path, "layout_image": output_layout_path }, "skipped": True }) return result_info # 创建临时会话目录 temp_dir, session_id = self.create_temp_session_dir() try: # 读取图片 image = Image.open(image_path) # 使用 DotsOCRParser 处理图片 filename = f"omnidocbench_{session_id}" results = self.parser.parse_image( input_path=image, filename=filename, prompt_mode=self.prompt_mode, save_dir=temp_dir, fitz_preprocess=True # 对图片使用 fitz 预处理 ) # 解析结果 if not results: raise Exception("未返回解析结果") result = results[0] # parse_image 返回单个结果的列表 # 保存所有结果文件到输出目录 saved_files = self.save_results_to_output_dir(result, image_name, output_dir) # 验证保存结果 success_count = sum(1 for path in saved_files.values() if path and os.path.exists(path)) if success_count >= 2: # 至少保存了 md 和 json result_info.update({ "success": True, "output_files": saved_files }) else: raise Exception(f"保存文件不完整 ({success_count}/3)") finally: # 清理临时目录 if os.path.exists(temp_dir): shutil.rmtree(temp_dir, ignore_errors=True) except Exception as e: result_info["error"] = str(e) print(f"❌ Error processing {image_name}: {e}") finally: result_info["processing_time"] = time.time() - start_time return result_info def process_images_single_process(image_paths: List[str], processor: DotsOCRProcessor, batch_size: int = 1, output_dir: str = "./output") -> List[Dict[str, Any]]: """ 单进程版本的图像处理函数 Args: image_paths: 图像路径列表 processor: DotsOCR处理器实例 batch_size: 批处理大小 output_dir: 输出目录 Returns: 处理结果列表 """ # 创建输出目录 output_path = Path(output_dir) output_path.mkdir(parents=True, exist_ok=True) all_results = [] total_images = len(image_paths) print(f"Processing {total_images} images with batch size {batch_size}") # 使用tqdm显示进度，添加更多统计信息 with tqdm(total=total_images, desc="Processing images", unit="img", bar_format='{l_bar}{bar}| {n_fmt}/{total_fmt} [{elapsed}<{remaining}, {rate_fmt}]') as pbar: # 按批次处理图像（DotsOCR通常单张处理） for i in range(0, total_images, batch_size): batch = image_paths[i:i + batch_size] batch_start_time = time.time() batch_results = [] try: # 处理批次中的每张图片 for image_path in batch: try: result = processor.process_single_image(image_path, output_dir) batch_results.append(result) except Exception as e: print(f"Error processing {image_path}: {e}", file=sys.stderr) traceback.print_exc() batch_results.append({ "image_path": image_path, "processing_time": 0, "success": False, "device": f"{processor.ip}:{processor.port}", "error": str(e) }) batch_processing_time = time.time() - batch_start_time all_results.extend(batch_results) # 更新进度条 success_count = sum(1 for r in batch_results if r.get('success', False)) skipped_count = sum(1 for r in batch_results if r.get('skipped', False)) total_success = sum(1 for r in all_results if r.get('success', False)) total_skipped = sum(1 for r in all_results if r.get('skipped', False)) avg_time = batch_processing_time / len(batch) pbar.update(len(batch)) pbar.set_postfix({ 'batch_time': f"{batch_processing_time:.2f}s", 'avg_time': f"{avg_time:.2f}s/img", 'success': f"{total_success}/{len(all_results)}", 'skipped': f"{total_skipped}", 'rate': f"{total_success/len(all_results)*100:.1f}%" }) except Exception as e: print(f"Error processing batch {[Path(p).name for p in batch]}: {e}", file=sys.stderr) traceback.print_exc() # 为批次中的所有图像添加错误结果 error_results = [] for img_path in batch: error_results.append({ "image_path": str(img_path), "processing_time": 0, "success": False, "device": f"{processor.ip}:{processor.port}", "error": str(e) }) all_results.extend(error_results) pbar.update(len(batch)) return all_results def process_images_concurrent(image_paths: List[str], processor: DotsOCRProcessor, batch_size: int = 1, output_dir: str = "./output", max_workers: int = 3) -> List[Dict[str, Any]]: """并发版本的图像处理函数""" from concurrent.futures import ThreadPoolExecutor, as_completed Path(output_dir).mkdir(parents=True, exist_ok=True) def process_batch(batch_images): """处理一批图像""" batch_results = [] for image_path in batch_images: try: result = processor.process_single_image(image_path, output_dir) batch_results.append(result) except Exception as e: batch_results.append({ "image_path": image_path, "processing_time": 0, "success": False, "device": f"{processor.ip}:{processor.port}", "error": str(e) }) return batch_results # 将图像分批 batches = [image_paths[i:i + batch_size] for i in range(0, len(image_paths), batch_size)] all_results = [] with ThreadPoolExecutor(max_workers=max_workers) as executor: # 提交所有批次 future_to_batch = {executor.submit(process_batch, batch): batch for batch in batches} # 使用 tqdm 显示进度 with tqdm(total=len(image_paths), desc="Processing images") as pbar: for future in as_completed(future_to_batch): try: batch_results = future.result() all_results.extend(batch_results) # 更新进度 success_count = sum(1 for r in batch_results if r.get('success', False)) pbar.update(len(batch_results)) pbar.set_postfix({'batch_success': f"{success_count}/{len(batch_results)}"}) except Exception as e: batch = future_to_batch[future] # 为批次中的所有图像添加错误结果 error_results = [ { "image_path": img_path, "processing_time": 0, "success": False, "device": f"{processor.ip}:{processor.port}", "error": str(e) } for img_path in batch ] all_results.extend(error_results) pbar.update(len(batch)) return all_results def main(): """主函数""" parser = argparse.ArgumentParser(description="DotsOCR Processing with PDF Support") # 输入参数组 input_group = parser.add_mutually_exclusive_group(required=True) input_group.add_argument("--input_file", type=str, help="Input file (supports both PDF and image file)") input_group.add_argument("--input_dir", type=str, help="Input directory (supports both PDF and image files)") input_group.add_argument("--input_file_list", type=str, help="Input file list (one file per line)") input_group.add_argument("--input_csv", type=str, help="Input CSV file with image_path and status columns") # 输出参数 parser.add_argument("--output_dir", type=str, help="Output directory") # DotsOCR 参数 parser.add_argument("--ip", type=str, default="127.0.0.1", help="vLLM server IP") parser.add_argument("--port", type=int, default=8101, help="vLLM server port") parser.add_argument("--model_name", type=str, default="DotsOCR", help="Model name") parser.add_argument("--prompt_mode", type=str, default="prompt_layout_all_en", choices=list(dict_promptmode_to_prompt.keys()), help="Prompt mode") parser.add_argument("--min_pixels", type=int, default=MIN_PIXELS, help="Minimum pixels") parser.add_argument("--max_pixels", type=int, default=MAX_PIXELS, help="Maximum pixels") parser.add_argument("--dpi", type=int, default=200, help="PDF processing DPI") parser.add_argument('--no-normalize', action='store_true', help='禁用数字标准化') # 处理参数 parser.add_argument("--batch_size", type=int, default=1, help="Batch size") parser.add_argument("--input_pattern", type=str, default="*", help="Input file pattern") parser.add_argument("--test_mode", action="store_true", help="Test mode (process only 10 images)") parser.add_argument("--collect_results", type=str, help="收集处理结果到指定CSV文件") # 并发参数 parser.add_argument("--max_workers", type=int, default=3, help="Maximum number of concurrent workers (should match vLLM data-parallel-size)") parser.add_argument("--use_threading", action="store_true", help="Use multi-threading") args = parser.parse_args() try: # 获取并预处理输入文件 print("🔄 Preprocessing input files...") image_files = get_input_files(args) if not image_files: print("❌ No input files found or processed") return 1 output_dir = Path(args.output_dir).resolve() print(f"📁 Output dir: {output_dir}") print(f"📊 Found {len(image_files)} image files to process") if args.test_mode: image_files = image_files[:10] print(f"🧪 Test mode: processing only {len(image_files)} images") print(f"🌐 Using server: {args.ip}:{args.port}") print(f"📦 Batch size: {args.batch_size}") print(f"🎯 Prompt mode: {args.prompt_mode}") # 创建处理器 processor = DotsOCRProcessor( ip=args.ip, port=args.port, model_name=args.model_name, prompt_mode=args.prompt_mode, dpi=args.dpi, min_pixels=args.min_pixels, max_pixels=args.max_pixels, normalize_numbers=not args.no_normalize ) # 开始处理 start_time = time.time() # 选择处理方式 if args.use_threading: results = process_images_concurrent( image_files, processor, args.batch_size, str(output_dir), args.max_workers ) else: results = process_images_single_process( image_files, processor, args.batch_size, str(output_dir) ) total_time = time.time() - start_time # 统计结果 success_count = sum(1 for r in results if r.get('success', False)) skipped_count = sum(1 for r in results if r.get('skipped', False)) error_count = len(results) - success_count pdf_page_count = sum(1 for r in results if r.get('is_pdf_page', False)) print(f"\n" + "="*60) print(f"✅ Processing completed!") print(f"📊 Statistics:") print(f" Total files processed: {len(image_files)}") print(f" PDF pages processed: {pdf_page_count}") print(f" Regular images processed: {len(image_files) - pdf_page_count}") print(f" Successful: {success_count}") print(f" Skipped: {skipped_count}") print(f" Failed: {error_count}") if len(image_files) > 0: print(f" Success rate: {success_count / len(image_files) * 100:.2f}%") print(f"⏱️ Performance:") print(f" Total time: {total_time:.2f} seconds") if total_time > 0: print(f" Throughput: {len(image_files) / total_time:.2f} images/second") print(f" Avg time per image: {total_time / len(image_files):.2f} seconds") # 保存结果统计 stats = { "total_files": len(image_files), "pdf_pages": pdf_page_count, "regular_images": len(image_files) - pdf_page_count, "success_count": success_count, "skipped_count": skipped_count, "error_count": error_count, "success_rate": success_count / len(image_files) if len(image_files) > 0 else 0, "total_time": total_time, "throughput": len(image_files) / total_time if total_time > 0 else 0, "avg_time_per_image": total_time / len(image_files) if len(image_files) > 0 else 0, "batch_size": args.batch_size, "server": f"{args.ip}:{args.port}", "model": args.model_name, "prompt_mode": args.prompt_mode, "pdf_dpi": args.dpi, "timestamp": time.strftime("%Y-%m-%d %H:%M:%S") } # 保存最终结果 output_file_name = Path(output_dir).name output_file = os.path.join(output_dir, f"{output_file_name}_results.json") final_results = { "stats": stats, "results": results } with open(output_file, 'w', encoding='utf-8') as f: json.dump(final_results, f, ensure_ascii=False, indent=2) print(f"💾 Results saved to: {output_file}") # 收集处理结果 if not args.collect_results: output_file_processed = Path(args.output_dir) / f"processed_files_{time.strftime('%Y%m%d_%H%M%S')}.csv" else: output_file_processed = Path(args.collect_results).resolve() processed_files = collect_pid_files(output_file) with open(output_file_processed, 'w', encoding='utf-8') as f: f.write("image_path,status\n") for file_path, status in processed_files: f.write(f"{file_path},{status}\n") print(f"💾 Processed files saved to: {output_file_processed}") return 0 except Exception as e: print(f"❌ Processing failed: {e}", file=sys.stderr) traceback.print_exc() return 1 if __name__ == "__main__": print(f"🚀 启动DotsOCR统一PDF/图像处理程序...") print(f"🔧 CUDA_VISIBLE_DEVICES: {os.environ.get('CUDA_VISIBLE_DEVICES', 'Not set')}") if len(sys.argv) == 1: # 如果没有命令行参数，使用默认配置运行 print("ℹ️ No command line arguments provided. Running with default configuration...") # 默认配置 default_config = { "input_file": "./sample_data/2023年度报告母公司_page_003.png", "output_dir": "./sample_data", "collect_results": "./sample_data/processed_files.csv", # "input_dir": "../../OmniDocBench/OpenDataLab___OmniDocBench/images", # "output_dir": "./OmniDocBench_DotsOCR_Results", # "collect_results": "./OmniDocBench_DotsOCR_Results/processed_files.csv", "ip": "10.192.72.11", # "ip": "127.0.0.1", "port": "8101", "model_name": "DotsOCR", "prompt_mode": "prompt_layout_all_en", "batch_size": "1", "max_workers": "3", "dpi": "200", } # 如果需要处理失败的文件，可以使用这个配置 # default_config = { # "input_csv": "./OmniDocBench_DotsOCR_Results/processed_files.csv", # "output_dir": "./OmniDocBench_DotsOCR_Results", # "ip": "127.0.0.1", # "port": "8101", # "collect_results": f"./OmniDocBench_DotsOCR_Results/processed_files_{time.strftime('%Y%m%d_%H%M%S')}.csv", # } # 构造参数 sys.argv = [sys.argv[0]] for key, value in default_config.items(): sys.argv.extend([f"--{key}", str(value)]) # 测试模式 sys.argv.append("--use_threading") # sys.argv.append("--test_mode") sys.exit(main())