Training Data Cleaning is a functional module provided by SuperMap iDesktopX for efficiently managing, editing, and preparing sample data for AI model training. It aims to help users systematically organize, classify, annotate, and correct large-scale image training datasets to improve data quality and lay a solid foundation for subsequent model training.

In the workflow of intelligent interpretation of remote sensing imagery, high-quality training data is key to model success. This function integrates scattered training data (images and corresponding label files) into a uniformly managed project through a visual interactive interface, bringing you the following key benefits:

• Systematic Management: Classify massive training data into positive and negative samples, modification status, etc., and store them structurally in the project for easy browsing and tracing.
• Quick Screening and Correction: Quickly classify samples using shortcut keys, and support visual drawing and editing of labels (such as adding positive sample areas, marking negative sample areas) to accurately correct erroneous annotations.
• Efficient Expansion and Export: Easily add new image data to expand the sample set, and export successfully processed samples by category into standard formats, seamlessly connecting to subsequent model training tasks.

In the following scenarios, you may use the Training Data Cleaning project.

• Before classification of ground objects/binary classification model training: Manually review and clean existing preliminary annotation results, remove erroneous samples, and correct imprecise label boundaries.
• Before detect change model training: Manage bitemporal image pairs and their labels to ensure the accuracy of change area annotations.
• When continuously iterating models: After the model produces new prediction results, import them as training data into this project for manual correction and confirmation, forming high-quality new training sets.

In this section, we will introduce the entire process operation of the Training Data Cleaning project:

• Basic Vocabulary: Introduces the core concepts and terminology involved in the function.

• New Project: Introduces how to create a new project.

• Open the project: Introduces how to load an existing project.

• Training Data Management: Introduces how to add new image data and export cleaned training data.
• Sample Classification and Label Editing: Introduces how to complete data cleaning through quick classification and visual annotation functions.
• Sample Label Style Settings: Introduces how to customize the display style of labels for a better visual editing experience.

Available starting from the SuperMap iDesktopX 2026 version.