Configuration

Q: What are the key hyperparameters for YOLO model accuracy?

Key hyperparameters affecting accuracy include: Adjust these based on your dataset and hardware. Learn more in Train Settings.

Q: How do I set the learning rate for training a YOLO model?

The learning rate (lr0) is crucial; start with 0.01 for SGD or 0.001 for Adam optimizer. Monitor metrics and adjust as needed. Use cosine learning rate schedulers (cos_lr) or warmup (warmup_epochs, warmup_momentum). Details are in the Train Guide.

Q: What are the default inference settings for YOLO models?

Default settings include: For a full overview, see Predict Settings and the Predict Guide.

Q: Why use mixed precision training with YOLO models?

Mixed precision training (amp=True) reduces memory usage and speeds up training using FP16 and FP32. It's beneficial for modern GPUs, allowing larger models and faster computations without significant accuracy loss. Learn more in the Train Guide.

YOLO settings and hyperparameters play a critical role in the model's performance, speed, and accuracy. These settings can affect the model's behavior at various stages, including training, validation, and prediction.

Watch: Mastering Ultralytics YOLO: Configuration

Ultralytics commands use the following syntax:

Example

CLIPython

yolo TASK MODE ARGS

from ultralytics import YOLO

# Load a YOLO model from a pre-trained weights file
model = YOLO("yolo11n.pt")

# Run MODE mode using the custom arguments ARGS (guess TASK)
model.MODE(ARGS)

Where:

TASK (optional) is one of (detect, segment, classify, pose, obb)
MODE (required) is one of (train, val, predict, export, track, benchmark)
ARGS (optional) are arg=value pairs like imgsz=640 that override defaults.

Default ARG values are defined on this page and come from the cfg/defaults.yaml file.

Tasks

Ultralytics YOLO models can perform a variety of computer vision tasks, including:

Detect: Object detection identifies and localizes objects within an image or video.
Segment: Instance segmentation divides an image or video into regions corresponding to different objects or classes.
Classify: Image classification predicts the class label of an input image.
Pose: Pose estimation identifies objects and estimates their keypoints in an image or video.
OBB: Oriented Bounding Boxes uses rotated bounding boxes, suitable for satellite or medical imagery.

Argument	Default	Description
`task`	`'detect'`	Specifies the YOLO task: `detect` for object detection, `segment` for segmentation, `classify` for classification, `pose` for pose estimation, and `obb` for oriented bounding boxes. Each task is tailored to specific outputs and problems in image and video analysis.

Tasks Guide

Modes

Ultralytics YOLO models operate in different modes, each designed for a specific stage of the model lifecycle:

Train: Train a YOLO model on a custom dataset.
Val: Validate a trained YOLO model.
Predict: Use a trained YOLO model to make predictions on new images or videos.
Export: Export a YOLO model for deployment.
Track: Track objects in real-time using a YOLO model.
Benchmark: Benchmark the speed and accuracy of YOLO exports (ONNX, TensorRT, etc.).

Argument	Default	Description
`mode`	`'train'`	Specifies the YOLO model's operating mode: `train` for model training, `val` for validation, `predict` for inference, `export` for converting to deployment formats, `track` for object tracking, and `benchmark` for performance evaluation. Each mode supports different stages, from development to deployment.

Modes Guide

Train Settings

Training settings for YOLO models include hyperparameters and configurations that affect the model's performance, speed, and accuracy. Key settings include batch size, learning rate, momentum, and weight decay. The choice of optimizer, loss function, and dataset composition also impact training. Tuning and experimentation are crucial for optimal performance. For more details, see the Ultralytics entrypoint function.

参数	类型	默认值	描述
`model`	`str`	`None`	指定用于训练的模型文件。接受一个路径，可以是`.pt`I need to translate the phrase "pretrained model or a" into Simplified Chinese while maintaining the original format.

预训练模型或一个.yaml配置文件。对于定义模型结构或初始化权重至关重要。 | data | str | NoneI notice that the content you've provided appears to be incomplete. It seems to be a fragment of a sentence or instruction about a path to a dataset configuration file, but it cuts off after "e.g.,".

Would you like me to translate this fragment into Simplified Chinese, or would you prefer to provide the complete text for translation?coco8.yaml这个文件包含特定于数据集的参数，包括训练路径和validation dataI need to translate the given text fragment to Simplified Chinese while preserving its format and meaning.

，类名和类的数量。 | | epochs | int | 100I'll translate the provided content into Simplified Chinese while maintaining the original format:

| 训练的总轮数。每个[epoch](https://www.ultralytics.com/glossary/epoch)表示对整个数据集的一次完整遍历。调整这个值可能会影响训练时间和模型性能。

）、60% GPU 内存利用率的自动模式（batch=-1I'll translate the given text into Simplified Chinese while preserving the original format and meaning:

Here's the translation:

), 在磁盘上 (diskI notice that the text you've provided appears to be a fragment rather than a complete sentence or paragraph. It seems to be part of a larger text describing options where something either "disables it" or some other action represented by the closing parenthesis at the beginning.

Without more context, I'll translate just this fragment as accurately as possible:

)，或禁用它 (False). 通过减少磁盘I/O来提高训练速度，但代价是增加内存使用量。 | device | int或者str或者list | None| 指定用于训练的计算设备：单个GPU (device=0），多个GPU（device=0,1I notice that the text you've provided is very short and appears to be a fragment. It looks like part of a technical description, possibly about hardware components, with "CPU" mentioned in parentheses.

Here's the translation to Simplified Chinese:

), CPU (

The text is identical in Simplified Chinese because it only contains the abbreviation "CPU" and some punctuation marks, which remain the same in both English and Simplified Chinese.device=cpuI'll translate the text to Simplified Chinese while maintaining the original format:

), 或者用于 Apple 芯片的 MPS (device=mps). | | workers | int | 8I need to translate the given text fragment into Simplified Chinese while preserving its original meaning and format.

  | 数据加载的工作线程数量（每`RANK`如果是多GPU训练）。影响数据预处理和输入模型的速度，尤其在多GPU设置中非常有用。

 | 禁用马赛克[data augmentation](https://www.ultralytics.com/glossary/data-augmentation)在最后的N个训练周期中稳定训练以便完成。设置为0将禁用此功能。

| SGD的动量因子或beta1为Adam optimizers，影响过去梯度在当前更新中的融合。 | weight_decay | float | 0.0005I need to translate "L2" to Simplified Chinese while maintaining the original format. However, "L2" appears to be a very brief code or label without much context. In technical contexts, "L2" often refers to "Layer 2" in networking or other technical domains, and it's typically kept as "L2" in Chinese technical documentation as well.

Given the minimal content and lack of context, I'll provide the most appropriate translation:

Note on Batch-size Settings

The batch argument offers three configuration options:

Fixed Batch Size: Specify the number of images per batch with an integer (e.g., batch=16).
Auto Mode (60% GPU Memory): Use batch=-1 for automatic adjustment to approximately 60% CUDA memory utilization.
Auto Mode with Utilization Fraction: Set a fraction (e.g., batch=0.70) to adjust based on a specified GPU memory usage.

Train Guide

Predict Settings

Prediction settings for YOLO models include hyperparameters and configurations that influence performance, speed, and accuracy during inference. Key settings include the confidence threshold, Non-Maximum Suppression (NMS) threshold, and the number of classes. Input data size, format, and supplementary features like masks also affect predictions. Tuning these settings is essential for optimal performance.

Inference arguments:

Argument	Type	Default	Description
`source`	`str`	`'ultralytics/assets'`	Specifies the data source for inference. Can be an image path, video file, directory, URL, or device ID for live feeds. Supports a wide range of formats and sources, enabling flexible application across different types of input.
`conf`	`float`	`0.25`	Sets the minimum confidence threshold for detections. Objects detected with confidence below this threshold will be disregarded. Adjusting this value can help reduce false positives.
`iou`	`float`	`0.7`	Intersection Over Union (IoU) threshold for Non-Maximum Suppression (NMS). Lower values result in fewer detections by eliminating overlapping boxes, useful for reducing duplicates.
`imgsz`	`int` or `tuple`	`640`	Defines the image size for inference. Can be a single integer `640` for square resizing or a (height, width) tuple. Proper sizing can improve detection accuracy and processing speed.
`half`	`bool`	`False`	Enables half-precision (FP16) inference, which can speed up model inference on supported GPUs with minimal impact on accuracy.
`device`	`str`	`None`	Specifies the device for inference (e.g., `cpu`, `cuda:0` or `0`). Allows users to select between CPU, a specific GPU, or other compute devices for model execution.
`batch`	`int`	`1`	Specifies the batch size for inference (only works when the source is a directory, video file or `.txt` file). A larger batch size can provide higher throughput, shortening the total amount of time required for inference.
`max_det`	`int`	`300`	Maximum number of detections allowed per image. Limits the total number of objects the model can detect in a single inference, preventing excessive outputs in dense scenes.
`vid_stride`	`int`	`1`	Frame stride for video inputs. Allows skipping frames in videos to speed up processing at the cost of temporal resolution. A value of 1 processes every frame, higher values skip frames.
`stream_buffer`	`bool`	`False`	Determines whether to queue incoming frames for video streams. If `False`, old frames get dropped to accommodate new frames (optimized for real-time applications). If `True`, queues new frames in a buffer, ensuring no frames get skipped, but will cause latency if inference FPS is lower than stream FPS.
`visualize`	`bool`	`False`	Activates visualization of model features during inference, providing insights into what the model is "seeing". Useful for debugging and model interpretation.
`augment`	`bool`	`False`	Enables test-time augmentation (TTA) for predictions, potentially improving detection robustness at the cost of inference speed.
`agnostic_nms`	`bool`	`False`	Enables class-agnostic Non-Maximum Suppression (NMS), which merges overlapping boxes of different classes. Useful in multi-class detection scenarios where class overlap is common.
`classes`	`list[int]`	`None`	Filters predictions to a set of class IDs. Only detections belonging to the specified classes will be returned. Useful for focusing on relevant objects in multi-class detection tasks.
`retina_masks`	`bool`	`False`	Returns high-resolution segmentation masks. The returned masks (`masks.data`) will match the original image size if enabled. If disabled, they have the image size used during inference.
`embed`	`list[int]`	`None`	Specifies the layers from which to extract feature vectors or embeddings. Useful for downstream tasks like clustering or similarity search.
`project`	`str`	`None`	Name of the project directory where prediction outputs are saved if `save` is enabled.
`name`	`str`	`None`	Name of the prediction run. Used for creating a subdirectory within the project folder, where prediction outputs are stored if `save` is enabled.
`stream`	`bool`	`False`	Enables memory-efficient processing for long videos or numerous images by returning a generator of Results objects instead of loading all frames into memory at once.
`verbose`	`bool`	`True`	Controls whether to display detailed inference logs in the terminal, providing real-time feedback on the prediction process.

Visualization arguments:

Argument	Type	Default	Description
`show`	`bool`	`False`	If `True`, displays the annotated images or videos in a window. Useful for immediate visual feedback during development or testing.
`save`	`bool`	`False or True`	Enables saving of the annotated images or videos to file. Useful for documentation, further analysis, or sharing results. Defaults to True when using CLI & False when used in Python.
`save_frames`	`bool`	`False`	When processing videos, saves individual frames as images. Useful for extracting specific frames or for detailed frame-by-frame analysis.
`save_txt`	`bool`	`False`	Saves detection results in a text file, following the format `[class] [x_center] [y_center] [width] [height] [confidence]`. Useful for integration with other analysis tools.
`save_conf`	`bool`	`False`	Includes confidence scores in the saved text files. Enhances the detail available for post-processing and analysis.
`save_crop`	`bool`	`False`	Saves cropped images of detections. Useful for dataset augmentation, analysis, or creating focused datasets for specific objects.
`show_labels`	`bool`	`True`	Displays labels for each detection in the visual output. Provides immediate understanding of detected objects.
`show_conf`	`bool`	`True`	Displays the confidence score for each detection alongside the label. Gives insight into the model's certainty for each detection.
`show_boxes`	`bool`	`True`	Draws bounding boxes around detected objects. Essential for visual identification and location of objects in images or video frames.
`line_width`	`None or int`	`None`	Specifies the line width of bounding boxes. If `None`, the line width is automatically adjusted based on the image size. Provides visual customization for clarity.
`font_size`	`float`	`None`	Text font size for annotations. Scales automatically with image size if set to `None`.
`font`	`str`	`'Arial.ttf'`	Font name or path for text annotations in the visualization.
`pil`	`bool`	`False`	Return image as a PIL Image object instead of numpy array.
`kpt_radius`	`int`	`5`	Radius of keypoints when visualizing pose estimation results.
`kpt_line`	`bool`	`True`	Connect keypoints with lines when visualizing pose estimation.
`masks`	`bool`	`True`	Display segmentation masks in the visualization output.
`probs`	`bool`	`True`	Include classification probabilities in the visualization.
`filename`	`str`	`None`	Path and filename to save the annotated image when `save=True`.
`color_mode`	`str`	`'class'`	Specify the coloring mode for visualizations, e.g., 'instance' or 'class'.
`txt_color`	`tuple[int, int, int]`	`(255, 255, 255)`	RGB text color for classification task annotations.

Predict Guide

Validation Settings

Validation settings for YOLO models involve hyperparameters and configurations to evaluate performance on a validation dataset. These settings influence performance, speed, and accuracy. Common settings include batch size, validation frequency, and performance metrics. The validation dataset's size and composition, along with the specific task, also affect the process.

Argument	Type	Default	Description
`data`	`str`	`None`	Specifies the path to the dataset configuration file (e.g., `coco8.yaml`). This file includes paths to validation data, class names, and number of classes.
`imgsz`	`int`	`640`	Defines the size of input images. All images are resized to this dimension before processing. Larger sizes may improve accuracy for small objects but increase computation time.
`batch`	`int`	`16`	Sets the number of images per batch. Higher values utilize GPU memory more efficiently but require more VRAM. Adjust based on available hardware resources.
`save_json`	`bool`	`False`	If `True`, saves the results to a JSON file for further analysis, integration with other tools, or submission to evaluation servers like COCO.
`conf`	`float`	`0.001`	Sets the minimum confidence threshold for detections. Lower values increase recall but may introduce more false positives. Used during validation to compute precision-recall curves.
`iou`	`float`	`0.6`	Sets the Intersection Over Union threshold for Non-Maximum Suppression. Controls duplicate detection elimination.
`max_det`	`int`	`300`	Limits the maximum number of detections per image. Useful in dense scenes to prevent excessive detections and manage computational resources.
`half`	`bool`	`True`	Enables half-precision (FP16) computation, reducing memory usage and potentially increasing speed with minimal impact on accuracy.
`device`	`str`	`None`	Specifies the device for validation (`cpu`, `cuda:0`, etc.). When `None`, automatically selects the best available device. Multiple CUDA devices can be specified with comma separation.
`dnn`	`bool`	`False`	If `True`, uses the OpenCV DNN module for ONNX model inference, offering an alternative to PyTorch inference methods.
`plots`	`bool`	`False`	When set to `True`, generates and saves plots of predictions versus ground truth, confusion matrices, and PR curves for visual evaluation of model performance.
`rect`	`bool`	`True`	If `True`, uses rectangular inference for batching, reducing padding and potentially increasing speed and efficiency by processing images in their original aspect ratio.
`split`	`str`	`'val'`	Determines the dataset split to use for validation (`val`, `test`, or `train`). Allows flexibility in choosing the data segment for performance evaluation.
`project`	`str`	`None`	Name of the project directory where validation outputs are saved. Helps organize results from different experiments or models.
`name`	`str`	`None`	Name of the validation run. Used for creating a subdirectory within the project folder, where validation logs and outputs are stored.
`verbose`	`bool`	`False`	If `True`, displays detailed information during the validation process, including per-class metrics, batch progress, and additional debugging information.
`save_txt`	`bool`	`False`	If `True`, saves detection results in text files, with one file per image, useful for further analysis, custom post-processing, or integration with other systems.
`save_conf`	`bool`	`False`	If `True`, includes confidence values in the saved text files when `save_txt` is enabled, providing more detailed output for analysis and filtering.
`save_crop`	`bool`	`False`	If `True`, saves cropped images of detected objects, which can be useful for creating focused datasets, visual verification, or further analysis of individual detections.
`workers`	`int`	`8`	Number of worker threads for data loading. Higher values can speed up data preprocessing but may increase CPU usage. Setting to 0 uses main thread, which can be more stable in some environments.
`augment`	`bool`	`False`	Enables test-time augmentation (TTA) during validation, potentially improving detection accuracy at the cost of inference speed by running inference on transformed versions of the input.
`agnostic_nms`	`bool`	`False`	Enables class-agnostic Non-Maximum Suppression, which merges overlapping boxes regardless of their predicted class. Useful for instance-focused applications.
`single_cls`	`bool`	`False`	Treats all classes as a single class during validation. Useful for evaluating model performance on binary detection tasks or when class distinctions aren't important.

Careful tuning and experimentation are crucial to ensure optimal performance and to detect and prevent overfitting.

Val Guide

Export Settings

Export settings for YOLO models include configurations for saving or exporting the model for use in different environments. These settings impact performance, size, and compatibility. Key settings include the exported file format (e.g., ONNX, TensorFlow SavedModel), the target device (e.g., CPU, GPU), and features like masks. The model's task and the destination environment's constraints also affect the export process.

Argument	Type	Default	Description
`format`	`str`	`'torchscript'`	Target format for the exported model, such as `'onnx'`, `'torchscript'`, `'engine'` (TensorRT), or others. Each format enables compatibility with different deployment environments.
`imgsz`	`int` or `tuple`	`640`	Desired image size for the model input. Can be an integer for square images (e.g., `640` for 640×640) or a tuple `(height, width)` for specific dimensions.
`keras`	`bool`	`False`	Enables export to Keras format for TensorFlow SavedModel, providing compatibility with TensorFlow serving and APIs.
`optimize`	`bool`	`False`	Applies optimization for mobile devices when exporting to TorchScript, potentially reducing model size and improving inference performance. Not compatible with NCNN format or CUDA devices.
`half`	`bool`	`False`	Enables FP16 (half-precision) quantization, reducing model size and potentially speeding up inference on supported hardware. Not compatible with INT8 quantization or CPU-only exports for ONNX.
`int8`	`bool`	`False`	Activates INT8 quantization, further compressing the model and speeding up inference with minimal accuracy loss, primarily for edge devices. When used with TensorRT, performs post-training quantization (PTQ).
`dynamic`	`bool`	`False`	Allows dynamic input sizes for ONNX, TensorRT and OpenVINO exports, enhancing flexibility in handling varying image dimensions. Automatically set to `True` when using TensorRT with INT8.
`simplify`	`bool`	`True`	Simplifies the model graph for ONNX exports with `onnxslim`, potentially improving performance and compatibility with inference engines.
`opset`	`int`	`None`	Specifies the ONNX opset version for compatibility with different ONNX parsers and runtimes. If not set, uses the latest supported version.
`workspace`	`float` or `None`	`None`	Sets the maximum workspace size in GiB for TensorRT optimizations, balancing memory usage and performance. Use `None` for auto-allocation by TensorRT up to device maximum.
`nms`	`bool`	`False`	Adds Non-Maximum Suppression (NMS) to the exported model when supported (see Export Formats), improving detection post-processing efficiency. Not available for end2end models.
`batch`	`int`	`1`	Specifies export model batch inference size or the maximum number of images the exported model will process concurrently in `predict` mode. For Edge TPU exports, this is automatically set to 1.
`device`	`str`	`None`	Specifies the device for exporting: GPU (`device=0`), CPU (`device=cpu`), MPS for Apple silicon (`device=mps`) or DLA for NVIDIA Jetson (`device=dla:0` or `device=dla:1`). TensorRT exports automatically use GPU.
`data`	`str`	`'coco8.yaml'`	Path to the dataset configuration file (default: `coco8.yaml`), essential for INT8 quantization calibration. If not specified with INT8 enabled, a default dataset will be assigned.
`fraction`	`float`	`1.0`	Specifies the fraction of the dataset to use for INT8 quantization calibration. Allows for calibrating on a subset of the full dataset, useful for experiments or when resources are limited. If not specified with INT8 enabled, the full dataset will be used.

Thoughtful configuration ensures the exported model is optimized for its use case and functions effectively in the target environment.

Export Guide

Solutions Settings

Ultralytics Solutions configuration settings offer flexibility to customize models for tasks like object counting, heatmap creation, workout tracking, data analysis, zone tracking, queue management, and region-based counting. These options allow easy adjustments for accurate and useful results tailored to specific needs.

Argument	Type	Default	Description
`model`	`str`	`None`	Path to Ultralytics YOLO Model File.
`region`	`list`	`[(20, 400), (1260, 400)]`	List of points defining the counting region.
`show_in`	`bool`	`True`	Flag to control whether to display the in counts on the video stream.
`show_out`	`bool`	`True`	Flag to control whether to display the out counts on the video stream.
`analytics_type`	`str`	`line`	Type of graph, i.e., `line`, `bar`, `area`, or `pie`.
`colormap`	`int`	`cv2.COLORMAP_JET`	Colormap to use for the heatmap.
`json_file`	`str`	`None`	Path to the JSON file that contains all parking coordinates data.
`up_angle`	`float`	`145.0`	Angle threshold for the 'up' pose.
`kpts`	`list[int, int, int]`	`[6, 8, 10]`	List of keypoints used for monitoring workouts. These keypoints correspond to body joints or parts, such as shoulders, elbows, and wrists, for exercises like push-ups, pull-ups, squats, ab-workouts.
`down_angle`	`float`	`90.0`	Angle threshold for the 'down' pose.
`blur_ratio`	`float`	`0.5`	Adjusts percentage of blur intensity, with values in range `0.1 - 1.0`.
`crop_dir`	`str`	`"cropped-detections"`	Directory name for storing cropped detections.
`records`	`int`	`5`	Total detections count to trigger an email with security alarm system.
`vision_point`	`tuple[int, int]`	`(50, 50)`	The point where vision will track objects and draw paths using VisionEye Solution.
`tracker`	`str`	`'botsort.yaml'`	Specifies the tracking algorithm to use, e.g., `bytetrack.yaml` or `botsort.yaml`.
`conf`	`float`	`0.3`	Sets the confidence threshold for detections; lower values allow more objects to be tracked but may include false positives.
`iou`	`float`	`0.5`	Sets the Intersection over Union (IoU) threshold for filtering overlapping detections.
`classes`	`list`	`None`	Filters results by class index. For example, `classes=[0, 2, 3]` only tracks the specified classes.
`verbose`	`bool`	`True`	Controls the display of tracking results, providing a visual output of tracked objects.
`device`	`str`	`None`	Specifies the device for inference (e.g., `cpu`, `cuda:0` or `0`). Allows users to select between CPU, a specific GPU, or other compute devices for model execution.
`show`	`bool`	`False`	If `True`, displays the annotated images or videos in a window. Useful for immediate visual feedback during development or testing.
`line_width`	`None or int`	`None`	Specifies the line width of bounding boxes. If `None`, the line width is automatically adjusted based on the image size. Provides visual customization for clarity.

Solutions Guide

Augmentation Settings

Data augmentation techniques are essential for improving YOLO model robustness and performance by introducing variability into the training data, helping the model generalize better to unseen data. The following table outlines each augmentation argument's purpose and effect:

参数	类型	默认值	范围	描述
`hsv_h`	`float`	`0.015`	`0.0 - 1.0`	通过色轮的一部分调整图像的色调，引入色彩变化。帮助模型在不同光照条件下实现泛化。
`hsv_s`	`float`	`0.7`	`0.0 - 1.0`	通过一定比例改变图像的饱和度，影响颜色的强度。适用于模拟不同的环境条件。
`hsv_v`	`float`	`0.4`	`0.0 - 1.0`	通过调整图像的亮度值，帮助模型在各种光照条件下都能表现良好。
`degrees`	`float`	`0.0`	`-180 - +180`	在指定的角度范围内随机旋转图像，提高模型识别各种方向上的物体的能力。
`translate`	`float`	`0.1`	`0.0 - 1.0`	水平和垂直方向上按图像尺寸的一部分平移图像，有助于学习检测部分可见的物体。
`scale`	`float`	`0.5`	`>=0.0`	通过增益因子缩放图像，模拟不同距离下的物体与相机的关系。
`shear`	`float`	`0.0`	`-180 - +180`	通过指定角度剪切图像，模拟从不同角度观看物体的效果。
`perspective`	`float`	`0.0`	`0.0 - 0.001`	对图像应用随机透视变换，增强模型理解三维空间中物体的能力。
`flipud`	`float`	`0.0`	`0.0 - 1.0`	以指定的概率将图像上下翻转，在不影响对象特征的情况下增加数据的多样性。
`fliplr`	`float`	`0.5`	`0.0 - 1.0`	以指定的概率将图像从左到右翻转，有助于学习对称物体和增加数据集多样性。
`bgr`	`float`	`0.0`	`0.0 - 1.0`	以指定的概率将图像通道从RGB翻转为BGR，有助于增强对不正确通道排序的鲁棒性。
`mosaic`	`float`	`1.0`	`0.0 - 1.0`	将四张训练图像合并为一张，模拟不同的场景组合和物体交互。对于复杂场景理解非常有效。
`mixup`	`float`	`0.0`	`0.0 - 1.0`	混合两张图像及其标签，创建一个合成图像。通过引入标签噪声和视觉变化，增强模型的泛化能力。
`copy_paste`	`float`	`0.0`	`0.0 - 1.0`	跨图像复制和粘贴对象，有助于增加对象实例数量并学习对象遮挡。需要分割标签。
`copy_paste_mode`	`str`	`'flip'`我将把这段内容翻译成简体中文，保持原始格式和含义：

    | -             | 在选项中选择复制粘贴增强方法 (`"flip"`, `"mixup"`).                                                                                      |

Adjust these settings to meet dataset and task requirements. Experimenting with different values can help find the optimal augmentation strategy for the best model performance.

Logging, Checkpoints and Plotting Settings

Logging, checkpoints, plotting, and file management are important when training a YOLO model:

Logging: Track the model's progress and diagnose issues using libraries like TensorBoard or by writing to a file.
Checkpoints: Save the model at regular intervals to resume training or experiment with different configurations.
Plotting: Visualize performance and training progress using libraries like matplotlib or TensorBoard.
File management: Organize files generated during training, such as checkpoints, log files, and plots, for easy access and analysis.

Effective management of these aspects helps track progress and makes debugging and optimization easier.

Argument	Default	Description
`project`	`'runs'`	Specifies the root directory for saving training runs. Each run is saved in a separate subdirectory.
`name`	`'exp'`	Defines the experiment name. If unspecified, YOLO increments this name for each run (e.g., `exp`, `exp2`) to avoid overwriting.
`exist_ok`	`False`	Determines whether to overwrite an existing experiment directory. `True` allows overwriting; `False` prevents it.
`plots`	`False`	Controls the generation and saving of training and validation plots. Set to `True` to create plots like loss curves, precision-recall curves, and sample predictions for visual tracking of performance.
`save`	`False`	Enables saving training checkpoints and final model weights. Set to `True` to save model states periodically, allowing training resumption or model deployment.

FAQ

How do I improve my YOLO model's performance during training?

Improve performance by tuning hyperparameters like batch size, learning rate, momentum, and weight decay. Adjust data augmentation settings, select the right optimizer, and use techniques like early stopping or mixed precision. For details, see the Train Guide.

What are the key hyperparameters for YOLO model accuracy?

Key hyperparameters affecting accuracy include:

Batch Size (batch): Larger sizes can stabilize training but need more memory.
Learning Rate (lr0): Smaller rates offer fine adjustments but slower convergence.
Momentum (momentum): Accelerates gradient vectors, dampening oscillations.
Image Size (imgsz): Larger sizes improve accuracy but increase computational load.

Adjust these based on your dataset and hardware. Learn more in Train Settings.

How do I set the learning rate for training a YOLO model?

The learning rate (lr0) is crucial; start with 0.01 for SGD or 0.001 for Adam optimizer. Monitor metrics and adjust as needed. Use cosine learning rate schedulers (cos_lr) or warmup (warmup_epochs, warmup_momentum). Details are in the Train Guide.

What are the default inference settings for YOLO models?

Default settings include:

Confidence Threshold (conf=0.25): Minimum confidence for detections.
IoU Threshold (iou=0.7): For Non-Maximum Suppression (NMS).
Image Size (imgsz=640): Resizes input images.
Device (device=None): Selects CPU or GPU.

For a full overview, see Predict Settings and the Predict Guide.

Why use mixed precision training with YOLO models?

Mixed precision training (amp=True) reduces memory usage and speeds up training using FP16 and FP32. It's beneficial for modern GPUs, allowing larger models and faster computations without significant accuracy loss. Learn more in the Train Guide.

📅 Created 1 year ago ✏️ Updated 26 days ago