yolo11n.yaml

### YOLOv11n 配置文件解析 #### 模型架构定义 YOLOv11 的配置文件位于 `ultralytics/cfg/models/11` 文件夹内，其中包含了不同任务类型的模型配置文件。对于目标检测任务而言，这些配置文件不仅规定了网络结构还设定了训练过程中的各项超参数。 ```yaml # Model Configuration File Example for Object Detection with YOLOv11n nc: 80 # number of classes (COCO dataset has 80 classes by default)[^1] depth_multiple: '0.33' # model depth multiple, scales the layers within backbone and neck modules width_multiple: '0.25' # model width multiple, affects channel numbers across all convolutional layers backbone: - [focus, [64, 3]] # Focus layer configuration; output channels=64, kernel size=3 - [conv, [128, 3, 2]] # Convolutional layer followed by maxpooling operation; output channels=128, kernel size=3, stride=2 ... neck: - [csp1, [128]] ... head: anchors: - [[10,13], [16,30], [33,23]] # anchor box dimensions at scale level P3 - [[30,61], [62,45], [59,119]] # anchor box dimensions at scale level P4 - [[116,90], [156,198], [373,326]] # anchor box dimensions at scale level P5 ``` 此段代码展示了 YOLOv11n 中典型的对象检测模型配置文件的一部分。它指明了类别的数量、宽度倍率、深度倍率等基本信息，并列举了一些骨干网层和颈部模块的具体设置。 #### 训练参数设定 除了上述提到的基础组件外，在实际应用过程中还需要调整一系列与训练有关的关键参数来优化最终效果： - **batch_size**: 批处理大小直接影响GPU内存占用情况及收敛速度。 - **epochs**: 总迭代次数决定了整个训练周期长度。 - **imgsz**: 输入图片尺寸影响特征提取效率及其分辨率质量。 ```yaml train: batch_size: 16 # Batch Size during training phase. epochs: 300 # Total number of epochs to train over. imgsz: 640 # Image size used as input into network. hyp: # Hyperparameters dictionary containing learning rate schedules etc. lr0: 0.01 # Initial Learning Rate. lrf: 0.1 # Final Learning Rate factor applied after cosine annealing scheduler completes its cycle. momentum: 0.937 # Momentum term in SGD optimizer formulation. weight_decay: 0.0005 # Weight decay coefficient added onto loss function calculation. ... ``` 以上部分摘自官方文档中关于如何配置训练阶段所需参数的内容[^3]。