YOLOv5全面解析教程⑦：使用模型融合提升mAP和mAR

撰文｜FengWen、BBuf

代码仓库地址：https://github.com/Oneflow-Inc/one-yolov5。欢迎star one-yolov5项目获取最新的动态。 如果你有问题，欢迎在仓库给我们提出宝贵的意见。如果对你有帮助，欢迎来给我Star呀~  

模型融合 (Model Ensembling)

融合建模（ Ensemble modeling） 是这样一个过程： 通过使用许多不同的建模算法或使用不同的训练数据集，创建多个不同的模型来预测结果。 然后，融合模型汇总每个基础 模型（ base model ） 的预测结果 ，并对未见过的数据进行 最终预测。 使用融合模型的动机是为了减少预测的泛化误差。 只要基础 模型具有多样性和独立性，采用融合方法时， 模型的预测误差就会减小。 这种方法在做出预测时寻求集体智慧（ wisdom of crowd ）。 尽管融合模型在模型中具有多个基础 模型，但它就像 单个模型那样 运行和执行。（https://www.sciencedirect.com/topics/computer-science/ensemble-modeling）

2

开始之前

  
  
  

   
   
   
git clone https://github.com/Oneflow-Inc/one-yolov5.gitcd one-yolov5pip install -r requirements.txt  # install
  
  
  
  
  
  

   
   
   
  
  
  

3

普通测试

  
  
  

   
   
   
$ python val.py --weights ./yolov5x --data coco.yaml --img 640 --half
  
  
  
  
  
  

   
   
   

  
  
  

  
  
  

   
   
   
val: data=data/coco.yaml, weights=['./yolov5x'], batch_size=32, imgsz=640, conf_thres=0.001, iou_thres=0.6, task=val, device=, workers=8, single_cls=False, augment=False, verbose=False, save_txt=False, save_hybrid=False, save_conf=False, save_json=True, project=runs/val, name=exp, exist_ok=False, half=True, dnn=FalseYOLOv5 🚀 v1.0-8-g94ec5c4 Python-3.8.13 oneflow-0.8.1.dev20221018+cu112 Fusing layers... Model summary: 322 layers, 86705005 parameters, 571965 gradientsval: Scanning '/data/dataset/fengwen/coco/val2017.cache' images and labels... 4952 found, 48 missing, 0 empty, 0 corrupt: 100%|████████               Class     Images     Labels          P          R     mAP@.5 mAP@.5:.95: 100%|██████████| 157/157 [01:55<00:00,  1.36it/                 all       5000      36335      0.743      0.627      0.685      0.503Speed: 0.1ms pre-process, 7.5ms inference, 2.1ms NMS per image at shape (32, 3, 640, 640)  # <--- baseline speed
Evaluating pycocotools mAP... saving runs/val/exp3/yolov5x_predictions.json...
...Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.505 # <--- baseline mAP Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.689 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.545 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.339 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.557 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.650 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.382 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.628 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.677  # <--- baseline mAR Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.523 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.730 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.826
  
  
  
  
  
  

   
   
   
  
  
  

4

融合测试

  
  
  

   
   
   
python val.py --weights ./yolov5x ./yolov5l6  --data data/coco.yaml --img 640 --half
  
  
  
  
  
  

   
   
   

  
  
  

  
  
  

   
   
   
val: data=data/coco.yaml, weights=['./yolov5x', './yolov5l6'], batch_size=32, imgsz=640, conf_thres=0.001, iou_thres=0.6, task=val, device=, workers=8, single_cls=False, augment=False, verbose=False, save_txt=False, save_hybrid=False, save_conf=False, save_json=True, project=runs/val, name=exp, exist_ok=False, half=True, dnn=FalseYOLOv5 🚀 v1.0-29-g8ed33f3 Python-3.8.13 oneflow-0.8.1.dev20221018+cu112 Fusing layers... Model summary: 322 layers, 86705005 parameters, 571965 gradientsFusing layers... Model summary: 346 layers, 76726332 parameters, 653820 gradientsEnsemble created with ['./yolov5x', './yolov5l6']
val: Scanning '/data/dataset/fengwen/coco/val2017.cache' images and labels... 4952 found, 48 missing, 0 empty, 0 corrupt: 100%|████████               Class     Images     Labels          P          R     mAP@.5 mAP@.5:.95: 100%|██████████| 157/157 [03:14<00:00,  1.24s/i                 all       5000      36335       0.73      0.644      0.693      0.513Speed: 0.1ms pre-process, 23.7ms inference, 2.3ms NMS per image at shape (32, 3, 640, 640) # <--- ensemble speed
Evaluating pycocotools mAP... saving runs/val/exp21/yolov5x_predictions.json...
... Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.515  # <--- ensemble mAP Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.697 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.556 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.340 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.567 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.678 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.389 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.637 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.690 # <--- ensemble mAR Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.517 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.743 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.842
  
  
  
  
  
  

   
   
   

  
  
  

5

融合推理

  
  
  

   
   
   
python detect.py --weights ./yolov5x ./yolov5l6 --img 640 --source  data/images
  
  
  
  
  
  

   
   
   

  
  
  

detect: weights=['./yolov5x', './yolov5l6'], source=data/images, data=data/coco128.yaml, imgsz=[640, 640], conf_thres=0.25, iou_thres=0.45, max_det=1000, device=, view_img=False, save_txt=False, save_conf=False, save_crop=False, nosave=False, classes=None, agnostic_nms=False, augment=False, visualize=False, update=False, project=runs/detect, name=exp, exist_ok=False, line_thickness=3, hide_labels=False, hide_conf=False, half=False, dnn=FalseYOLOv5 🚀 v1.0-31-g6b1387c Python-3.8.13 oneflow-0.8.1.dev20221018+cu112 Fusing layers... Model summary: 322 layers, 86705005 parameters, 571965 gradientsFusing layers... Model summary: 346 layers, 76726332 parameters, 653820 gradientsEnsemble created with ['./yolov5x', './yolov5l6']
detect.py:159: DeprecationWarning: In future, it will be an error for 'np.bool_' scalars to be interpreted as an index  s += f"{n} {names[int(c)]}{'s' * (n > 1)}, "  # add to stringimage 1/2 /home/fengwen/one-yolov5/data/images/bus.jpg: 640x512 4 persons, 1 bus, 1 handbag, 1 tie, Done. (0.028s)detect.py:159: DeprecationWarning: In future, it will be an error for 'np.bool_' scalars to be interpreted as an index  s += f"{n} {names[int(c)]}{'s' * (n > 1)}, "  # add to stringimage 2/2 /home/fengwen/one-yolov5/data/images/zidane.jpg: 384x640 3 persons, 2 ties, Done. (0.023s)0.6ms pre-process, 25.6ms inference, 2.4ms NMS per image at shape (1, 3, 640, 640)

参考文章

• https://github.com/ultralytics/yolov5/issues/318

• Transformer模型的基础演算
  

• 关于大型语言模型的争论和局限
  

• 揭秘RLHF；可商用开源LLM列表

• John Schulman：通往TruthGPT之路

• 为什么ChatGPT用强化学习而非监督学习

• OneEmbedding:单卡 训练TB级推荐模型不是梦
• GLM训练加速：性能最高提升3倍，显存节省1/3

试用OneFlow: github.com/Oneflow-Inc/oneflow/