#!/usr/bin/env python3 # -*- coding: utf-8 -*- """ 测试并行OCR性能,对比优化前后的提升 """ import os import sys import time from concurrent.futures import ThreadPoolExecutor from PIL import Image import numpy as np def test_serial_ocr(image_path: str, num_tasks: int = 3): """测试串行OCR(优化前的方式)""" print("=" * 70) print("测试1: 串行OCR(优化前)") print("=" * 70) print(f"任务数量: {num_tasks}") print(f"OMP_NUM_THREADS: {os.environ.get('OMP_NUM_THREADS', '未设置')}") print() # 加载图片 try: img = Image.open(image_path) if img.mode != 'RGB': img = img.convert('RGB') img_np = np.array(img) print(f"图片尺寸: {img.size[0]}x{img.size[1]}") print() except Exception as e: print(f"[ERROR] 无法加载图片: {e}") return None # 初始化OCR try: from operations.ocr_operations import _ocr_processor _ocr_processor._ensure_ocr() if _ocr_processor.ocr is None: print("[ERROR] OCR引擎未初始化") return None print(f"OCR引擎: {_ocr_processor.ocr_type}") print() except Exception as e: print(f"[ERROR] OCR引擎初始化失败: {e}") return None # 串行执行OCR任务 start_time = time.time() results = [] print("开始串行OCR测试...") for i in range(num_tasks): task_start = time.time() try: if _ocr_processor.ocr_type == "easyocr": result = _ocr_processor.ocr.readtext(img_np) else: result = _ocr_processor.ocr.ocr(img_np) # 提取文本 from Underlying_Operations import underlying_operations ops = underlying_operations() text = ops.extract_ocr_content(result) task_elapsed = (time.time() - task_start) * 1000 results.append(text) print(f" 任务 {i+1}/{num_tasks}: {task_elapsed:.0f}ms, 结果: {text[:30] if text else '(空)'}") except Exception as e: print(f" 任务 {i+1}/{num_tasks}: 失败 - {e}") results.append(None) total_elapsed = (time.time() - start_time) * 1000 print() print(f"总耗时: {total_elapsed:.0f}ms ({total_elapsed/1000:.2f}秒)") print(f"平均每个任务: {total_elapsed/num_tasks:.0f}ms") print() return total_elapsed def test_parallel_ocr(image_path: str, num_tasks: int = 3): """测试并行OCR(优化后的方式)""" print("=" * 70) print("测试2: 并行OCR(优化后)") print("=" * 70) print(f"任务数量: {num_tasks}") print(f"OMP_NUM_THREADS: {os.environ.get('OMP_NUM_THREADS', '未设置')}") # 获取最优线程数 try: from operations.screenshot_service import get_optimal_workers optimal_workers = get_optimal_workers() print(f"线程池大小: {optimal_workers}") except: optimal_workers = 6 print(f"线程池大小: {optimal_workers} (默认)") print() # 加载图片 try: img = Image.open(image_path) if img.mode != 'RGB': img = img.convert('RGB') img_np = np.array(img) print(f"图片尺寸: {img.size[0]}x{img.size[1]}") print() except Exception as e: print(f"[ERROR] 无法加载图片: {e}") return None # 初始化OCR try: from operations.ocr_operations import _ocr_processor _ocr_processor._ensure_ocr() if _ocr_processor.ocr is None: print("[ERROR] OCR引擎未初始化") return None print(f"OCR引擎: {_ocr_processor.ocr_type}") print() except Exception as e: print(f"[ERROR] OCR引擎初始化失败: {e}") return None # 定义OCR任务函数 def ocr_task(task_id): """单个OCR任务""" try: if _ocr_processor.ocr_type == "easyocr": result = _ocr_processor.ocr.readtext(img_np) else: result = _ocr_processor.ocr.ocr(img_np) # 提取文本 from Underlying_Operations import underlying_operations ops = underlying_operations() text = ops.extract_ocr_content(result) return text except Exception as e: print(f" 任务 {task_id} 失败: {e}") return None # 并行执行OCR任务 start_time = time.time() results = [] print("开始并行OCR测试...") with ThreadPoolExecutor(max_workers=optimal_workers) as executor: # 提交所有任务 futures = {executor.submit(ocr_task, i+1): i+1 for i in range(num_tasks)} # 收集结果 task_times = {} for future in futures: task_id = futures[future] task_start = time.time() try: result = future.result() task_elapsed = (time.time() - task_start) * 1000 task_times[task_id] = task_elapsed results.append(result) print(f" 任务 {task_id}/{num_tasks}: {task_elapsed:.0f}ms, 结果: {result[:30] if result else '(空)'}") except Exception as e: print(f" 任务 {task_id}/{num_tasks}: 失败 - {e}") results.append(None) total_elapsed = (time.time() - start_time) * 1000 print() print(f"总耗时: {total_elapsed:.0f}ms ({total_elapsed/1000:.2f}秒)") if task_times: avg_task_time = sum(task_times.values()) / len(task_times) print(f"平均每个任务: {avg_task_time:.0f}ms") print(f"并行效率: {avg_task_time * num_tasks / total_elapsed:.1f}x (理想情况下应该接近{num_tasks}x)") print() return total_elapsed def test_parallel_ocr_old_config(image_path: str, num_tasks: int = 3): """测试并行OCR(使用旧的配置:OMP_NUM_THREADS=1, max_workers=6)""" print("=" * 70) print("测试3: 并行OCR(旧配置:OMP=1, 线程池=6)") print("=" * 70) print(f"任务数量: {num_tasks}") # 临时设置旧配置 old_omp = os.environ.get('OMP_NUM_THREADS', '4') os.environ['OMP_NUM_THREADS'] = '1' print(f"OMP_NUM_THREADS: 1 (临时设置)") print(f"线程池大小: 6 (固定)") print() # 加载图片 try: img = Image.open(image_path) if img.mode != 'RGB': img = img.convert('RGB') img_np = np.array(img) print(f"图片尺寸: {img.size[0]}x{img.size[1]}") print() except Exception as e: print(f"[ERROR] 无法加载图片: {e}") os.environ['OMP_NUM_THREADS'] = old_omp return None # 重新初始化OCR(使用新的OMP设置) try: from operations.ocr_operations import _ocr_processor # 注意:OCR引擎已经初始化,OMP设置可能不会立即生效 # 但我们可以测试线程池的效果 print(f"OCR引擎: {_ocr_processor.ocr_type}") print("注意: OMP设置可能需要重启程序才能完全生效") print() except Exception as e: print(f"[ERROR] OCR引擎检查失败: {e}") os.environ['OMP_NUM_THREADS'] = old_omp return None # 定义OCR任务函数 def ocr_task(task_id): """单个OCR任务""" try: if _ocr_processor.ocr_type == "easyocr": result = _ocr_processor.ocr.readtext(img_np) else: result = _ocr_processor.ocr.ocr(img_np) # 提取文本 from Underlying_Operations import underlying_operations ops = underlying_operations() text = ops.extract_ocr_content(result) return text except Exception as e: print(f" 任务 {task_id} 失败: {e}") return None # 并行执行OCR任务(使用旧的线程池大小) start_time = time.time() results = [] print("开始并行OCR测试(旧配置)...") with ThreadPoolExecutor(max_workers=6) as executor: # 提交所有任务 futures = {executor.submit(ocr_task, i+1): i+1 for i in range(num_tasks)} # 收集结果 task_times = {} for future in futures: task_id = futures[future] task_start = time.time() try: result = future.result() task_elapsed = (time.time() - task_start) * 1000 task_times[task_id] = task_elapsed results.append(result) print(f" 任务 {task_id}/{num_tasks}: {task_elapsed:.0f}ms, 结果: {result[:30] if result else '(空)'}") except Exception as e: print(f" 任务 {task_id}/{num_tasks}: 失败 - {e}") results.append(None) total_elapsed = (time.time() - start_time) * 1000 # 恢复OMP设置 os.environ['OMP_NUM_THREADS'] = old_omp print() print(f"总耗时: {total_elapsed:.0f}ms ({total_elapsed/1000:.2f}秒)") if task_times: avg_task_time = sum(task_times.values()) / len(task_times) print(f"平均每个任务: {avg_task_time:.0f}ms") print() return total_elapsed def main(): """主函数""" # 默认图片路径 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" # 从命令行参数获取图片路径 if len(sys.argv) > 1: image_path = sys.argv[1] else: image_path = default_image # 转换为绝对路径 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) # 检查文件是否存在 if not os.path.exists(image_path): print(f"[ERROR] 图片文件不存在: {image_path}") return # 获取任务数量(默认3个,减少测试时间) num_tasks = 3 if len(sys.argv) > 2: try: num_tasks = int(sys.argv[2]) except: pass print("=" * 70) print("并行OCR性能对比测试") print("=" * 70) print() print(f"测试图片: {os.path.basename(image_path)}") print(f"任务数量: {num_tasks}") print() # 获取当前配置信息 cpu_count = os.cpu_count() or 4 omp_threads = os.environ.get('OMP_NUM_THREADS', '未设置') print(f"系统信息:") print(f" CPU核心数: {cpu_count}") print(f" OMP_NUM_THREADS: {omp_threads}") 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: pass print() print() # 测试1: 串行OCR serial_time = test_serial_ocr(image_path, num_tasks) print() print() # 测试2: 并行OCR(优化后) parallel_time = test_parallel_ocr(image_path, num_tasks) print() print() # 测试3: 并行OCR(旧配置) parallel_old_time = test_parallel_ocr_old_config(image_path, num_tasks) # 性能对比总结 print() print("=" * 70) print("性能对比总结") print("=" * 70) print() if serial_time and parallel_time: speedup = serial_time / parallel_time print(f"串行OCR: {serial_time:.0f}ms ({serial_time/1000:.2f}秒)") print(f"并行OCR: {parallel_time:.0f}ms ({parallel_time/1000:.2f}秒)") print(f"性能提升: {speedup:.2f}x ({(1 - parallel_time/serial_time)*100:.1f}% 时间减少)") print() if parallel_old_time and parallel_time: speedup_vs_old = parallel_old_time / parallel_time print(f"旧配置并行: {parallel_old_time:.0f}ms ({parallel_old_time/1000:.2f}秒)") print(f"新配置并行: {parallel_time:.0f}ms ({parallel_time/1000:.2f}秒)") print(f"配置优化提升: {speedup_vs_old:.2f}x ({(1 - parallel_time/parallel_old_time)*100:.1f}% 时间减少)") print() if serial_time and parallel_time and parallel_old_time: total_speedup = serial_time / parallel_time print(f"总体优化效果:") print(f" 串行 → 并行: {serial_time/parallel_time:.2f}x") print(f" 旧配置 → 新配置: {parallel_old_time/parallel_time:.2f}x") print(f" 串行 → 新配置: {total_speedup:.2f}x") print() print("=" * 70) print("测试完成") print("=" * 70) if __name__ == "__main__": main()