使用真实值训练 AI 模型
已标注的真实值数据对于在各种计算机视觉任务（包括目标检测、语义分割、图像分类和视频活动识别）中训练有监督 AI 模型至关重要。Computer Vision Toolbox™ 提供了工具来帮助您准备已标注的真实值数据，用于深度学习训练，包括选择相关标签，修改文件路径，合并真实值目标以及组织训练和评估数据集。
图像标注器和视频标注器以 groundTruth 对象的形式导出已标注的真实值数据。要通过将已标注的真实值转换为与 AI 模型兼容的格式来生成训练数据集，可以使用 objectDetectorTrainingData、pixelLabelTrainingData 和 sceneLabelTrainingData 等函数。这些函数支持目标检测、分割和分类任务。有关详细信息，请参阅Training Data for Object Detection and Semantic Segmentation和Postprocess Exported Labels for Instance Segmentation Training。您还可以使用 polyToBlockedImage 创建分块图像表示，从而高效处理大规模图像。
要从真实值数据中选择特定标签，并根据任务要求过滤和组织注解，可以使用 selectLabelsByGroup、selectLabelsByType 或 selectLabelsByName 等函数。工具箱还支持使用以下函数对已标注数据进行后处理：merge 用于合并多个真实值目标，changeFilePaths 用于更新数据集引用，以及 gatherLabelData 用于提取标签信息。对于视频数据，writeVideoScenes 和 sceneTimeRanges 等实用工具可帮助管理场景级注解。
要与同事共享和审核图像标签，请考虑在图像标注器中创建团队项目。有关详细信息，请参阅Get Started with Team-Based Labeling。
函数
主题
- Elements of Ground Truth Objects
Understand how to save and pass data using a ground truth data object.
- Share and Store Labeled Ground Truth Data
Share and store labeled ground truth data exported from labeling apps.
- How Labeler Apps Store Exported Pixel Labels
Learn how the labeling apps store pixel label data.
- Training Data for Object Detection and Semantic Segmentation
Create training data for object detection or semantic segmentation using the Image Labeler or Video Labeler.
- Postprocess Exported Labels for Instance Segmentation Training
Postprocess exported ground truth labels and create training datastore for training instance segmentation networks such as SOLOv2 or Mask R-CNN.
- Datastores for Deep Learning (Deep Learning Toolbox)
Learn how to use datastores in deep learning applications.
- Get Started with Image Preprocessing and Augmentation for Deep Learning
Preprocess data for deep learning applications with deterministic operations such as resizing, or augment training data with randomized operations such as random cropping.
精选示例
Create Instance Segmentation Training Data From Ground Truth
Create instance segmentation training data from a groundTruth object. To train a Mask R-CNN and SOLO v2 network using the trainMaskRCNN and trainSOLOv2 functions, respectively, format your input data as a 1-by-4 cell array containing the RGB training image, bounding boxes, instance labels, and instance masks. Convert your polygon ground data into a set of binary instance masks and axis-aligned rectangles for training.
