YOLOv11 目标检测

指定用于训练的模型文件。接受 .pt 预训练模型或 .yaml 配置文件的路径。对于定义模型结构或初始化权重至关重要。

Specifies the model file for training. Accepts a path to either a .pt pretrained model or a .yaml configuration file. Essential for defining the model structure or initializing weights.

数据集配置文件的路径（例如，coco8.yaml）。该文件包含特定于数据集的参数，包括训练和验证数据的路径、类名和类数。

Path to the dataset configuration file (e.g., coco8.yaml). This file contains dataset-specific parameters, including paths to training and validation data, class names, and number of classes.

​训练 epoch 的总数。每个 epoch 表示整个数据集的完整传递。调整此值可能会影响训练持续时间和模型性能。

Total number of training epochs. Each epoch represents a full pass over the entire dataset. Adjusting this value can affect training duration and model performance.

最大训练时间（小时）。如果设置，这将覆盖 epochs 参数，从而允许训练在指定的持续时间后自动停止。适用于时间受限的训练场景。

Maximum training time in hours. If set, this overrides the epochs argument, allowing training to automatically stop after the specified duration. Useful for time-constrained training scenarios.

​在提前停止训练之前，在验证指标没有改善的情况下等待的epochs数。通过在性能停滞时停止训练来帮助防止过度拟合。

Number of epochs to wait without improvement in validation metrics before early stopping the training. Helps prevent overfitting by stopping training when performance plateaus.

具有三种模式：设置为整数（例如，batch=16）、GPU 内存利用率为 60% 的自动模式 （batch=-1） 或具有指定利用率分数的自动模式 （batch=0.70）。

Batch size, with three modes: set as an integer (e.g., batch=16), auto mode for 60% GPU memory utilization (batch=-1), or auto mode with specified utilization fraction (batch=0.70).

用于训练的目标图像大小。在输入到模型之前，所有图像都会调整到此尺寸。影响模型精度和计算复杂度。

Target image size for training. All images are resized to this dimension before being fed into the model. Affects model accuracy and computational complexity.

支持保存训练检查点和最终模型权重。对于恢复训练或模型部署非常有用。

Enables saving of training checkpoints and final model weights. Useful for resuming training or model deployment.

保存模型检查点的频率，以 epoch 为单位指定。值为 -1 将禁用此功能。对于在长时间训练期间保存临时模型很有用。

Frequency of saving model checkpoints, specified in epochs. A value of -1 disables this feature. Useful for saving interim models during long training sessions.

启用数据集图像在内存 （True/ram）、磁盘 （disk） 上的缓存，或禁用数据集图像缓存 （False）。通过减少磁盘 I/O 来提高训练速度，但代价是内存使用量增加。

Enables caching of dataset images in memory (True/ram), on disk (disk), or disables it (False). Improves training speed by reducing disk I/O at the cost of increased memory usage.

指定用于训练的计算设备：单个 GPU （device=0）、多个 GPU （device=0,1）、CPU （device=cpu） 或 Apple Silicon 的 MPS （device=mps）。

Specifies the computational device(s) for training: a single GPU (device=0), multiple GPUs (device=0,1), CPU (device=cpu), or MPS for Apple silicon (device=mps).

用于数据加载的工作线程数（如果是多 GPU 训练，则按 RANK 计算）。影响数据预处理和馈送到模型的速度，在多 GPU 设置中特别有用。

Number of worker threads for data loading (per RANK if Multi-GPU training). Influences the speed of data preprocessing and feeding into the model, especially useful in multi-GPU setups.

保存训练输出的项目目录的名称。允许有序地存储不同的实验。

Name of the project directory where training outputs are saved. Allows for organized storage of different experiments.

训练运行的名称。用于在项目文件夹中创建子目录，用于存储训练日志和输出。

Name of the training run. Used for creating a subdirectory within the project folder, where training logs and outputs are stored.

如果为 True，则允许覆盖现有 project/name 目录。对于迭代实验很有用，无需手动清除以前的输出。

If True, allows overwriting of an existing project/name directory. Useful for iterative experimentation without needing to manually clear previous outputs.

确定是否从预训练的模型开始训练。可以是布尔值或指向要从中加载权重的特定模型的字符串路径。提高训练效率和模型性能。

Determines whether to start training from a pretrained model. Can be a boolean value or a string path to a specific model from which to load weights. Enhances training efficiency and model performance.

选择用于训练的优化器。选项包括 SGD、Adam、AdamW、NAdam、RAdam、RMSProp 等，或 auto，用于根据模型配置自动选择。影响收敛速度和稳定性。

Choice of optimizer for training. Options include SGD, Adam, AdamW, NAdam, RAdam, RMSProp etc., or auto for automatic selection based on model configuration. Affects convergence speed and stability.

设置用于训练的随机种子，确保结果在具有相同配置的运行中的可重复性。

Sets the random seed for training, ensuring reproducibility of results across runs with the same configurations.

强制使用确定性算法，确保可重现性，但由于对非确定性算法的限制，可能会影响性能和速度。

Forces deterministic algorithm use, ensuring reproducibility but may affect performance and speed due to the restriction on non-deterministic algorithms.

在训练期间，将多类数据集中的所有类视为单个类。对于二元分类任务或专注于对象存在而不是分类时非常有用。

Treats all classes in multi-class datasets as a single class during training. Useful for binary classification tasks or when focusing on object presence rather than classification.

指定要训练的类 ID 列表。对于在训练期间筛选并仅关注某些类很有用。

Specifies a list of class IDs to train on. Useful for filtering out and focusing only on certain classes during training.

启用矩形训练，优化批量合成以实现最少的填充。可以提高效率和速度，但可能会影响模型准确性。

Enables rectangular training, optimizing batch composition for minimal padding. Can improve efficiency and speed but may affect model accuracy.

通过在训练期间将 imgsz 增加/减少 0.5 倍来实现多尺度训练。在推理过程中使用多个 imgsz 训练模型以使其更准确。

Enables multi-scale training by increasing/decreasing imgsz by upto a factor of 0.5 during training. Trains the model to be more accurate with multiple imgsz during inference.

利用余弦学习速率调度器，在历元上按照余弦曲线调整学习速率。有助于管理学习率以实现更好的收敛。

Utilizes a cosine learning rate scheduler, adjusting the learning rate following a cosine curve over epochs. Helps in managing learning rate for better convergence.

从上次保存的检查点恢复训练。自动加载模型权重、优化器状态和 epoch 计数，无缝继续训练。

Resumes training from the last saved checkpoint. Automatically loads model weights, optimizer state, and epoch count, continuing training seamlessly.

支持自动混合精度（AMP）训练，减少内存使用，并可能加快训练速度，同时对准确性的影响最小。

Enables Automatic Mixed Precision (AMP) training, reducing memory usage and possibly speeding up training with minimal impact on accuracy.

指定用于训练的数据集的分数。允许对完整数据集的子集进行训练，这对于实验或资源有限时非常有用。

Specifies the fraction of the dataset to use for training. Allows for training on a subset of the full dataset, useful for experiments or when resources are limited.

在训练期间启用 ONNX 和 TensorRT 速度分析，有助于优化模型部署。

Enables profiling of ONNX and TensorRT speeds during training, useful for optimizing model deployment.

冻结模型的前N层或按索引指定的层，减少可训练参数的数量。有助于微调或迁移学习。

Freezes the first N layers of the model or specified layers by index, reducing the number of trainable parameters. Useful for fine-tuning or transfer learning.

初始学习率（即 SGD=1E-2，Adam=1E-3）。调整此值对于优化过程至关重要，它会影响模型权重的更新速度。

Initial learning rate (i.e. SGD=1E-2, Adam=1E-3) . Adjusting this value is crucial for the optimization process, influencing how rapidly model weights are updated.

最终学习率占初始学习率的分数 = （lr0 * lrf），与调度器结合使用，以随时间调整学习率。

Final learning rate as a fraction of the initial rate = (lr0 * lrf), used in conjunction with schedulers to adjust the learning rate over time.

​SGD 的动量因子或 Adam 优化器的 beta1，影响当前更新中过去梯度的合并。

Momentum factor for SGD or beta1 for Adam optimizers, influencing the incorporation of past gradients in the current update.

L2正则化项，惩罚大权重以防止过度拟合。

L2 regularization term, penalizing large weights to prevent overfitting.

学习率预热的 epoch 数，逐渐将学习率从低值提高到初始学习率，以尽早稳定训练。

Number of epochs for learning rate warmup, gradually increasing the learning rate from a low value to the initial learning rate to stabilize training early on.

预热阶段的初始动量，在预热期间逐渐适应设定的动量。

Initial momentum for warmup phase, gradually adjusting to the set momentum over the warmup period.

预热阶段偏差参数的学习率，有助于稳定初始 epoch 中的模型训练。

Learning rate for bias parameters during the warmup phase, helping stabilize model training in the initial epochs.

损失函数中 box loss 分量的权重，影响对准确预测边界框坐标的重视程度。

Weight of the box loss component in the loss function, influencing how much emphasis is placed on accurately predicting bounding box coordinates.

分类损失在总损失函数中的权重，影响相对于其他分量的正确类别预测的重要性。

Weight of the classification loss in the total loss function, affecting the importance of correct class prediction relative to other components.

分布焦点损失的权重，用于某些 YOLO 版本以进行精细分类。

Weight of the distribution focal loss, used in certain YOLO versions for fine-grained classification.

为姿势估计训练的模型中姿势损失的权重，影响对准确预测姿势关键点的重视。

Weight of the pose loss in models trained for pose estimation, influencing the emphasis on accurately predicting pose keypoints.

姿态估计模型中关键点客体损失的权重，平衡检测置信度和姿态准确性。

Weight of the keypoint objectness loss in pose estimation models, balancing detection confidence with pose accuracy.

损失标准化的标称批量。

Nominal batch size for normalization of loss.

确定是应将对象掩码合并到单个掩码中进行训练，还是为每个对象保持单独。在重叠的情况下，在合并过程中，较小的蒙版将覆盖在较大的蒙版之上。

Determines whether object masks should be merged into a single mask for training, or kept separate for each object. In case of overlap, the smaller mask is overlaid on top of the larger mask during merge.

分割掩码的下采样率，影响训练期间使用的掩码的分辨率。

Downsample ratio for segmentation masks, affecting the resolution of masks used during training.

分类任务中正则化的 dropout rate，防止在训练期间通过随机省略单元来过度拟合。

Dropout rate for regularization in classification tasks, preventing overfitting by randomly omitting units during training.

在训练期间启用验证，允许在单独的数据集上定期评估模型性能。

Enables validation during training, allowing for periodic evaluation of model performance on a separate dataset.

生成并保存训练和验证指标的图以及预测示例，提供对模型性能和学习进度的可视化见解。

Generates and saves plots of training and validation metrics, as well as prediction examples, providing visual insights into model performance and learning progression.