前言

最近在研究如何让YOLOv5推理得更快，总体看来，主要有以下这些思路：

• 使用更快的 GPU，即：P100 -> V100 -> A100
• 多卡GPU推理
• 减小模型尺寸，即YOLOv5x -> YOLOv5l -> YOLOv5m -> YOLOv5s -> YOLOv5n
• 进行半精度FP16推理与python detect.py --half
• 减少–img-size，即 1280 -> 640 -> 320
• 导出成ONNX或OpenVINO格式，获得CPU加速
• 导出到TensorRT获得GPU加速
• 批量输入图片进行推理
• 使用多进程/多线程进行推理

注：使用多卡GPU和多进程/多线程的推理并不会对单张图片推理起到加速作用，只适用于很多张图片一起进行推理的场景。

本篇主要来研究多进程/多线程是否能对YOLOv5算法推理起到加速作用。

实验环境

GPU：RTX2060
torch：1.7.1+cu110
检测图片大小：1920×1080
img-size：1920
使用半精度推理half=True
推理模型：yolov5m.pt

实验过程

先放实验代码(detect.py)，根据官方源码进行了小改：

import configparserimport timefrom pathlib import Pathimport cv2import torchimport threadingimport sysimport multiprocessing as mpsys.path.append("yolov5")from models.experimental import attempt_loadfrom utils.datasets import LoadImagesfrom utils.general import check_img_size, non_max_suppression, scale_coordsfrom utils.plots import Annotator, colorsfrom utils.torch_utils import select_devicefrom concurrent.futures import ThreadPoolExecutorDetect_path = 'D:/Data/detect_outputs'# 检测图片输出路径def detect(path, model_path, detect_size):source = pathweights = model_pathimgsz = detect_sizeconf_thres = 0.25iou_thres = 0.45device = ""augment = Truesave_img = Truesave_dir = Path(Detect_path)# increment rundevice = select_device(device)half = device.type != 'cpu'# half precision only supported on CUDA# Load modelmodel = attempt_load(weights, map_location=device)# load FP32 modelstride = int(model.stride.max())# model strideimgsz = check_img_size(imgsz, s=stride)# check img_sizefif half:model.half()# to FP16# Set Dataloadervid_path, vid_writer = None, Nonedataset = LoadImages(source, img_size=imgsz, stride=stride)# Get names and colorsnames = model.module.names if hasattr(model, 'module') else model.names# Run inferenceif device.type != 'cpu':model(torch.zeros(1, 3, imgsz, imgsz).to(device).type_as(next(model.parameters())))# run onceresult_list = []for path, img, im0s, vid_cap in dataset:# 读取图片传到gpu上t1 = time.time()img = torch.from_numpy(img).to(device)print("read pictures cost time：", time.time() - t1)t2 = time.time()img = img.half() if half else img.float()# uint8 to fp16/32img /= 255.0# 0 - 255 to 0.0 - 1.0if img.ndimension() == 3:img = img.unsqueeze(0)print("process pictures cost time：", time.time() - t2)# Inferencepred = model(img, augment=augment)[0]# Apply NMSpred = non_max_suppression(pred, conf_thres, iou_thres)# Process detectionsfor i, det in enumerate(pred):# detections per imagep, s, im0, frame = path, '', im0s, getattr(dataset, 'frame', 0)p = Path(p)# to Pathsave_path = str(save_dir / p.name)# img.jpgs += '%gx%g ' % img.shape[2:]# print string# print(s)# 384x640s_result = ''# 输出检测结果annotator = Annotator(im0, line_width=3, example=str(names))if len(det):# Rescale boxes from img_size to im0 sizedet[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()# Print resultsfor c in det[:, -1].unique():n = (det[:, -1] == c).sum()# detections per class# s += f"{n} {names[int(c)]}{'s' * (n > 1)}, "# add to strings += f"{n} {names[int(c)]}, "# add to strings_result += f"{n} {names[int(c)]} "# Write resultsfor *xyxy, conf, cls in reversed(det):if save_img:c = int(cls)# label = f'{names[int(cls)]} {conf:.2f}'label = f'{names[int(cls)]}'# print(label)annotator.box_label(xyxy, label, color=colors(c, True))# print(xyxy)print(f'{s}')# print(f'{s_result}')result_list.append(s_result)# 将conf对象中的数据写入到文件中conf = configparser.ConfigParser()cfg_file = open("glovar.cfg", 'w')conf.add_section("default")# 在配置文件中增加一个段# 第一个参数是段名，第二个参数是选项名，第三个参数是选项对应的值conf.set("default", "process", str(dataset.img_count))conf.set("default", "total", str(dataset.nf))conf.write(cfg_file)cfg_file.close()im0 = annotator.result()# Save results (image with detections)t3 = time.time()if save_img:if dataset.mode == 'image':cv2.imwrite(save_path, im0)else:# 'video' or 'stream'if vid_path != save_path:# new videovid_path = save_pathif isinstance(vid_writer, cv2.VideoWriter):vid_writer.release()# release previous video writerif vid_cap:# videofps = vid_cap.get(cv2.CAP_PROP_FPS)w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))else:# streamfps, w, h = 30, im0.shape[1], im0.shape[0]save_path += '.mp4'vid_writer = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))vid_writer.write(im0)print("write pictures cost time：", time.time() - t3)print('Done')def run(path, model_path, detect_size):with torch.no_grad():detect(path, model_path, detect_size)

首先进行小批量的图片进行实验，下面输入两张图片进行检测。

原始推理

if __name__ == '__main__':s_t = time.time()path1 = "D:/Data/image/DJI_0001_00100.jpg"path2 = "D:/Data/image/DJI_0001_00530.jpg"model_path = "../weights/best.pt"detect_size = 1920run(path1, model_path, detect_size)run(path2, model_path, detect_size)print("Tatal Cost Time:", time.time() - s_t)

Tatal Cost Time: 3.496427059173584

线程池推理

开辟两个线程进行推理：

if __name__ == '__main__':s_t = time.time()pool = ThreadPoolExecutor(max_workers=2)path1 = "D:/Data/image/DJI_0001_00100.jpg"path2 = "D:/Data/image/DJI_0001_00530.jpg"model_path = "../weights/best.pt"detect_size = 1920pool.submit(run, path1, model_path, detect_size)pool.submit(run, path2, model_path, detect_size)pool.shutdown(wait=True)print("Tatal Cost Time:", time.time() - s_t)

Tatal Cost Time: 3.2433135509490967

开双线程推理和原始推理时间类似，再次验证了python中的”伪多线程”。

进程池推理

开辟两个进程进行推理：

if __name__ == '__main__':s_t = time.time()pool = mp.Pool(processes=2)path1 = "D:/Data/image/DJI_0001_00100.jpg"path2 = "D:/Data/image/DJI_0001_00530.jpg"model_path = "../weights/best.pt"detect_size = 1920pool.apply_async(run, (path1, model_path, detect_size,))pool.apply_async(run, (path2, model_path, detect_size,))pool.close()pool.join()print("Tatal Cost Time:", time.time() - s_t)

Tatal Cost Time: 6.020772695541382

双进程推理

双进程推理时间竟然是原始推理的两倍，以为是进程池的开销太大，于是换种写法，不使用进程池：

if __name__ == '__main__':s_t = time.time()path1 = "D:/Data/image/DJI_0001_00100.jpg"path2 = "D:/Data/image/DJI_0001_00530.jpg"model_path = "../weights/best.pt"detect_size = 1920p1 = mp.Process(target=run, args=(path1, model_path, detect_size,))p2 = mp.Process(target=run, args=(path2, model_path, detect_size,))p1.start()p2.start()p1.join()p2.join()print("Tatal Cost Time:", time.time() - s_t)

Tatal Cost Time: 6.089479446411133

发现双进程时间仍然较久，说明在数据较少时，进程的开销成本过高，这和我之前做的实验多线程和多进程的效率对比结果相类似。

于是下面将图像数量扩大到300张进行实验。

300pic-原始推理

if __name__ == '__main__':s_t = time.time()path1 = "D:/Data/image"path2 = "D:/Data/image2"path3 = "D:/Data/image3"model_path = "../weights/best.pt"detect_size = 1920run(path1, model_path, detect_size)run(path2, model_path, detect_size)run(path3, model_path, detect_size)print("Tatal Cost Time:", time.time() - s_t)

Tatal Cost Time: 62.02898120880127

300pic-多进程推理

if __name__ == '__main__':s_t = time.time()path1 = "D:/Data/image"path2 = "D:/Data/image2"path3 = "D:/Data/image3"model_path = "../weights/best.pt"detect_size = 1920p1 = mp.Process(target=run, args=(path1, model_path, detect_size,))p2 = mp.Process(target=run, args=(path2, model_path, detect_size,))p3 = mp.Process(target=run, args=(path3, model_path, detect_size,))p1.start()p2.start()p3.start()p1.join()p2.join()p3.join()print("Tatal Cost Time:", time.time() - s_t)

Tatal Cost Time: 47.85872721672058

和预期一样，当数据量提升上去时，多进程推理的速度逐渐超越原始推理。

总结

本次实验结果如下表所示：

值得注意的是，使用多进程推理时，进程间保持独立，这意味着模型需要被重复在GPU上进行创建，因此，可以根据单进程所占显存大小来估算显卡所支持的最大进程数。

后续：在顶配机上进行实验

后面嫖到了组里i9-13700K+RTX4090的顶配主机，再进行实验，结果如下：

后记：更正结论

后面觉得之前做的实验有些草率，尽管Python存在GIL的限制，但是在此类IO频繁的场景中，多线程仍然能缓解IO阻塞，从而实现加速，因此选用YOLOv5s模型，在4090上，对不同分辨率的图片进行测试：

输入图像分辨率：1920×1080

输入图像分辨率：13400×9528