写论文做消融实验的时候，间隔期可能浪费时间。

根目录下新建一个文件mytrain.py

from ultralytics import YOLO

if __name__ == '__main__':
'''
    如果是3个模块（A, B, C）要进行测试，排列组合数是3+2+1=7次实验+1次baseline=8次
    如果考虑模块插入部位的变化，组合数剧增。
'''
#TODO 实验0：baseline
    model = YOLO('yolov8s.yaml')
    model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s)

#TODO 实验1：+module1
    model = YOLO('yolov8s+module1.yaml')
    model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module1)

#TODO 实验2：+module2
    model = YOLO('yolov8s+module2.yaml')
    model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module2)

#TODO 实验3：+module3
    model = YOLO('yolov8s+module3.yaml')
    model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module3)

#TODO 实验4：+module1+module2
    model = YOLO('yolov8s+module1+module2.yaml')
    model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module1+module2)

#TODO 实验5：+module1+module3
    model = YOLO('yolov8s+module1+module3.yaml')
    model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module1+module3)

#TODO 实验6：+module2+module3
    model = YOLO('yolov8s+module2+module3.yaml')
    model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module2+module3)

#TODO 实验7：+module1+module2+module3
    model = YOLO('yolov8s+module1+module2+module3.yaml')
    model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module1+module2+module3)

运行后可以自动依次往下执行，如果中间某个yaml文件配置出错，就会中断，所有最好提前全部测试一遍，保证无人值守的情况下顺利执行全部实验。算一算每个epoch的所需时间，就可以估计几天后再见了！

于是先把model.train()注释掉。

from ultralytics import YOLO

if __name__ == '__main__':
'''
    如果是3个模块（A, B, C）要进行测试，排列组合数是3+2+1=7次实验+1次baseline=8次
    如果考虑模块插入部位的变化，组合数剧增。
'''
#TODO 实验0：baseline
    model = YOLO('yolov8s.yaml')
    # model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s)

#TODO 实验1：+module1
    model = YOLO('yolov8s+module1.yaml')
    # model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module1)

#TODO 实验2：+module2
    model = YOLO('yolov8s+module2.yaml')
    # model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module2)

#TODO 实验3：+module3
    model = YOLO('yolov8s+module2.yaml')
    # model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module3)

#TODO 实验4：+module1+module2
    model = YOLO('yolov8s+module1+module2.yaml')
    # model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module1+module2)

#TODO 实验5：+module1+module3
    model = YOLO('yolov8s+module1+module3.yaml')
    # model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module1+module3)

#TODO 实验6：+module2+module3
    model = YOLO('yolov8s+module2+module3.yaml')
    # model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module2+module3)

#TODO 实验7：+module1+module2+module3
    model = YOLO('yolov8s+module1+module2+module3.yaml')
    # model.train(data='mydataset.yaml', epochs=300, imgsz=640, batch=16, workers=0, pretrained=False, save_dir=runs\detect\yolov8s+module1+module2+module3)

训练结束之后，再批量验证一下

from ultralytics import YOLO

if __name__ == '__main__':

    model = YOLO('yolov8s.yaml').load('runs/detect/yolov8s/weights/best.pt')
    model.val(split='test',data='mydataset.yaml', imgsz=640, batch=1)

    model = YOLO('yolov8s+module1.yaml').load('runs/detect/yolov8s+module1/weights/best.pt')
    model.val(split='test',data='mydataset.yaml', imgsz=640, batch=1)

    model = YOLO('yolov8s+module2.yaml').load('runs/detect/yolov8s+module2/weights/best.pt')
    model.val(split='test',data='mydataset.yaml', imgsz=640, batch=1)

    model = YOLO('yolov8s+module3.yaml').load('runs/detect/yolov8s+module3/weights/best.pt')
    model.val(split='test',data='mydataset.yaml', imgsz=640, batch=1)

    model = YOLO('yolov8s+module1+module2.yaml').load('runs/detect/yolov8s+module1+module2/weights/best.pt')
    model.val(split='test',data='mydataset.yaml', imgsz=640, batch=1)

    model = YOLO('yolov8s+module1+module3.yaml').load('runs/detect/yolov8s+module1+module3/weights/best.pt')
    model.val(split='test',data='mydataset.yaml', imgsz=640, batch=1)

    model = YOLO('yolov8s+module2+module3.yaml').load('runs/detect/yolov8s+module2+module3/weights/best.pt')
    model.val(split='test',data='mydataset.yaml', imgsz=640, batch=1)

    model = YOLO('yolov8s+module1+module2+module3.yaml').load('runs/detect/yolov8s+module1+module2+module3/weights/best.pt')
    model.val(split='test',data='mydataset.yaml', imgsz=640, batch=1)