#!/usr/bin/env python3 # -*- coding: utf-8 -*- """ 真实场景并行OCR性能对比测试 模拟实际使用中的多区域并行OCR识别 """ import os import sys import time from concurrent.futures import ThreadPoolExecutor from PIL import Image import numpy as np import tempfile def create_test_regions(full_image_path): """从完整截图中创建多个测试区域(模拟真实OCR场景)""" try: img = Image.open(full_image_path) if img.mode != 'RGB': img = img.convert('RGB') width, height = img.size # 判断是否是完整截图(1080x2400)还是裁剪区域 is_full_screenshot = width >= 1000 and height >= 2000 if is_full_screenshot: # 完整截图:创建多个测试区域 # 模拟真实的OCR区域(基于福袋详情页的常见区域) regions = { "福袋倒计时_分钟": (400, 1050, 500, 1150), # 倒计时分钟区域 "福袋倒计时_秒": (480, 1110, 570, 1230), # 倒计时秒区域(用户提供的区域) "福袋内容": (50, 800, 500, 1000), # 福袋内容区域 "福袋价值": (200, 1300, 400, 1400), # 价值区域 "任务列表": (50, 600, 500, 750), # 任务列表区域 "参与按钮": (300, 1500, 800, 1600), # 按钮状态区域 } else: # 裁剪区域:使用当前图片作为测试区域之一,创建其他相似大小的区域 # 从文件名提取坐标信息(如果存在) coords_from_filename = None if "(480,1110)-(570,1230)" in full_image_path: coords_from_filename = (480, 1110, 570, 1230) # 创建多个相似大小的测试区域 # 如果图片较小,创建多个相同大小的区域用于测试 # 这些区域会在完整截图的合理位置 regions = {} # 使用当前图片作为第一个区域 if coords_from_filename: x1, y1, x2, y2 = coords_from_filename regions["福袋倒计时_秒"] = (x1, y1, x2, y2) # 创建其他相邻区域 offset_x = x2 - x1 offset_y = y2 - y1 regions["福袋倒计时_分钟"] = (x1 - offset_x - 20, y1, x1 - 20, y2) regions["福袋内容"] = (x1, y1 - offset_y - 20, x2, y1 - 20) regions["福袋价值"] = (x1, y2 + 20, x2, y2 + offset_y + 20) else: # 如果无法从文件名提取坐标,使用图片本身作为唯一测试区域 # 创建多个相同大小的区域(通过复制当前图片) regions["区域1"] = (0, 0, width, height) regions["区域2"] = (0, 0, width, height) # 相同区域,用于测试并行 regions["区域3"] = (0, 0, width, height) # 相同区域,用于测试并行 regions["区域4"] = (0, 0, width, height) # 相同区域,用于测试并行 regions["区域5"] = (0, 0, width, height) # 相同区域,用于测试并行 regions["区域6"] = (0, 0, width, height) # 相同区域,用于测试并行 # 裁剪并保存区域图片 region_images = {} temp_dir = tempfile.gettempdir() for name, (x1, y1, x2, y2) in regions.items(): try: # 确保坐标在图片范围内 x1 = max(0, min(x1, img.width)) y1 = max(0, min(y1, img.height)) x2 = max(x1, min(x2, img.width)) y2 = max(y1, min(y2, img.height)) if x2 > x1 and y2 > y1: crop_img = img.crop((x1, y1, x2, y2)) temp_path = os.path.join(temp_dir, f"ocr_test_{name}.png") crop_img.save(temp_path) region_images[name] = { 'path': temp_path, 'size': crop_img.size, 'coords': (x1, y1, x2, y2) } except (ValueError, OSError, IOError) as e: print(f"[WARNING] 无法裁剪区域 {name}: {e}") return region_images, img.size except (OSError, IOError, ValueError) as e: print(f"[ERROR] 无法处理图片: {e}") return {}, None def ocr_single_region(image_path, region_name, ocr_processor): """对单个区域进行OCR识别""" try: img = Image.open(image_path) if img.mode != 'RGB': img = img.convert('RGB') img_np = np.array(img) start_time = time.time() if ocr_processor.ocr_type == "easyocr": result = ocr_processor.ocr.readtext(img_np) else: result = ocr_processor.ocr.ocr(img_np) elapsed = (time.time() - start_time) * 1000 # 提取文本 from Underlying_Operations import underlying_operations ops = underlying_operations() text = ops.extract_ocr_content(result) return { 'region': region_name, 'text': text, 'time': elapsed, 'success': True } except Exception as e: import traceback error_msg = f"{type(e).__name__}: {str(e)}" print(f" [DEBUG] {region_name} OCR错误: {error_msg}") return { 'region': region_name, 'text': '', 'time': 0, 'success': False, 'error': error_msg } def test_parallel_ocr_real(region_images, config_name, omp_threads, max_workers): """测试真实场景的并行OCR""" print(f"\n{'='*70}") print(f"测试: {config_name}") print(f"{'='*70}") print(f"配置: OMP_NUM_THREADS={omp_threads}, max_workers={max_workers}") print(f"区域数量: {len(region_images)}") print() # 设置OMP线程数 old_omp = os.environ.get('OMP_NUM_THREADS', '4') os.environ['OMP_NUM_THREADS'] = str(omp_threads) # 初始化OCR(使用新的OMP设置) try: from operations.ocr_operations import OCRProcessor ocr_processor = OCRProcessor() # 访问受保护方法以初始化OCR引擎(测试代码中需要) # 使用 getattr 来避免 pylance 警告 ensure_ocr = getattr(ocr_processor, '_ensure_ocr') ensure_ocr() if ocr_processor.ocr is None: print("[ERROR] OCR引擎未初始化") os.environ['OMP_NUM_THREADS'] = old_omp return None print(f"OCR引擎: {ocr_processor.ocr_type}") print() except (ImportError, AttributeError, RuntimeError) as e: print(f"[ERROR] OCR引擎初始化失败: {e}") os.environ['OMP_NUM_THREADS'] = old_omp return None # 显示区域信息 print("OCR区域信息:") for name, info in region_images.items(): print(f" {name}: {info['size'][0]}x{info['size'][1]} 像素, 坐标: {info['coords']}") print() # 并行执行OCR print("开始并行OCR识别...") start_time = time.time() results = [] def ocr_task(name, path): return ocr_single_region(path, name, ocr_processor) with ThreadPoolExecutor(max_workers=max_workers) as executor: # 提交所有任务 futures = { executor.submit(ocr_task, name, info['path']): name for name, info in region_images.items() } # 收集结果 for future in futures: region_name = futures[future] try: result = future.result(timeout=60) # 每个任务最多60秒 results.append(result) status = "✓" if result['success'] else "✗" print(f" {status} {region_name}: {result['time']:.0f}ms, 结果: {result['text'][:30] if result['text'] else '(空)'}") except TimeoutError: print(f" ✗ {region_name}: 超时(>60秒)") results.append({ 'region': region_name, 'text': '', 'time': 0, 'success': False, 'error': 'Timeout' }) except (RuntimeError, AttributeError, KeyError) as e: print(f" ✗ {region_name}: 失败 - {e}") results.append({ 'region': region_name, 'text': '', 'time': 0, 'success': False, 'error': str(e) }) total_elapsed = (time.time() - start_time) * 1000 # 恢复OMP设置 os.environ['OMP_NUM_THREADS'] = old_omp # 统计结果 successful_results = [r for r in results if r['success']] total_task_time = sum(r['time'] for r in successful_results) avg_task_time = total_task_time / len(successful_results) if successful_results else 0 print() print(f"总耗时: {total_elapsed:.0f}ms ({total_elapsed/1000:.2f}秒)") print(f"任务总耗时: {total_task_time:.0f}ms") print(f"平均每个任务: {avg_task_time:.0f}ms") if total_task_time > 0: efficiency = total_task_time / total_elapsed print(f"并行效率: {efficiency:.2f}x (理想值: {len(region_images):.1f}x)") print() return { 'total_time': total_elapsed, 'task_times': [r['time'] for r in successful_results], 'results': results, 'config': config_name } def main(): """主函数""" # 默认图片路径(完整截图) default_image = "pic/debug_captures/2025-12-31/20251231_152116_ocr_福袋倒计时_秒_预处理后_(480,1110)-(570,1230)_模式_1_尺寸_270x360_来源_福袋倒计时_秒.png_有完整截图_是_有坐标_是_类型_digits.png" if len(sys.argv) > 1: image_path = sys.argv[1] else: image_path = default_image if not os.path.isabs(image_path): project_root = os.path.dirname(os.path.abspath(__file__)) image_path = os.path.join(project_root, image_path) if not os.path.exists(image_path): print(f"[ERROR] 图片文件不存在: {image_path}") return print("=" * 70) print("真实场景并行OCR性能对比测试") print("=" * 70) print() print(f"测试图片: {os.path.basename(image_path)}") print() # 创建测试区域 print("准备测试区域...") region_images, full_size = create_test_regions(image_path) if not region_images: print("[ERROR] 无法创建测试区域") return if full_size is not None: print(f"完整截图尺寸: {full_size[0]}x{full_size[1]}") else: print("完整截图尺寸: 未知") print(f"成功创建 {len(region_images)} 个测试区域") print() # 获取系统信息 cpu_count = os.cpu_count() or 4 print("系统信息:") print(f" CPU核心数: {cpu_count}") try: from operations.screenshot_service import get_optimal_workers optimal_workers = get_optimal_workers() print(f" 推荐线程池大小: {optimal_workers}") except (ImportError, AttributeError): optimal_workers = 6 print() # 测试1: 优化前配置(OMP=1, workers=6) result_old = test_parallel_ocr_real( region_images, "优化前配置", omp_threads=1, max_workers=6 ) # 等待一下,让系统稳定 time.sleep(1) # 测试2: 优化后配置(OMP=4, workers=10) result_new = test_parallel_ocr_real( region_images, "优化后配置", omp_threads=4, max_workers=optimal_workers ) # 性能对比总结 print() print("=" * 70) print("性能对比总结") print("=" * 70) print() if result_old and result_new: old_time = result_old['total_time'] new_time = result_new['total_time'] speedup = old_time / new_time if new_time > 0 else 0 print(f"优化前总耗时: {old_time:.0f}ms ({old_time/1000:.2f}秒)") print(f"优化后总耗时: {new_time:.0f}ms ({new_time/1000:.2f}秒)") print(f"性能提升: {speedup:.2f}x") print(f"时间减少: {(1 - new_time/old_time)*100:.1f}%") print() # 详细对比 if result_old['task_times'] and result_new['task_times']: old_avg = sum(result_old['task_times']) / len(result_old['task_times']) new_avg = sum(result_new['task_times']) / len(result_new['task_times']) print("任务级性能对比:") print(f" 优化前平均任务耗时: {old_avg:.0f}ms") print(f" 优化后平均任务耗时: {new_avg:.0f}ms") print(f" 单任务性能提升: {old_avg/new_avg:.2f}x") print() # 识别结果对比 print("识别结果对比:") for old_result in result_old['results']: region_name = old_result['region'] new_result = next((r for r in result_new['results'] if r['region'] == region_name), None) if old_result['success'] and new_result and new_result['success']: old_text = old_result['text'][:20] if old_result['text'] else '(空)' new_text = new_result['text'][:20] if new_result['text'] else '(空)' match = "✓" if old_text == new_text else "✗" print(f" {match} {region_name}: 结果一致") # 清理临时文件 print() print("清理临时文件...") for info in region_images.values(): try: if os.path.exists(info['path']): os.remove(info['path']) except (OSError, PermissionError): pass print() print("=" * 70) print("测试完成") print("=" * 70) if __name__ == "__main__": main()