Samples Category
Feature Description
The Samples Category group in the Training Data Cleaning tab provides a set of efficient shortcuts and buttons, allowing users to quickly and batch classify the status of the currently browsed samples. This operation automatically determines whether the label matches the classification category. For example, if marked as all positive samples, the system will check if the label contains pixels other than all positive samples. If it does, the label is modified to all positive samples; if not, no modification is made. It is suitable for macro quality judgment and filtering of entire samples.
SuperMap iDesktopX Available starting from the 2026 version.
Instructions
- Select the target sample: In the Training Data Manager, expand the Unclassified group and select a sample record. The map will synchronously display the image and label of this sample.
- Perform the classification operation: Click the corresponding button in the Samples Category group, or use its shortcut key, to mark the current sample as the corresponding category. The provided function descriptions are as follows:
- All Positive Samples [A]: Mark the current sample as a completely correct positive sample. Suitable for high-quality positive samples where all targets in the entire image are correctly labeled with no omissions.
- All Negative Samples [S]: Mark the current sample as a negative sample (background or no targets). Suitable for samples that contain no targets in the entire image, or serve as pure background samples.
- Delete [D]: Move the current sample to the Deleted group. Suitable for samples with extremely poor labeling quality, unusable, or erroneous samples.
- Do Not Modify [F]: Keep the current state of the sample without changes. Suitable for samples that have been reviewed and deemed unnecessary to modify, explicitly marking them to avoid repeated inspection.
- View the result: After the operation, the sample record will move from the Unclassified group to the corresponding All Positive Samples group, and this operation will be recorded in the *.sda.xml file, without being recorded in the Modified folder of the storage path. At the same time, the next record in the Unclassified group will be selected in the Training Data Manager.
Label Editing
Feature Description
The Label Editing group in the Training Data Cleaning tab provides a set of vector drawing tools, allowing users to visually modify sample labels, including adding positive sample areas, subtracting negative sample areas, and adjusting area shapes. This operation generates a new label file and is a core method for microscopic correction of samples.
When you find local errors in a sample's label (such as over-labeling, missing labeling, or inaccurate boundaries), you may use this function.
SuperMap iDesktopX Available starting from the 2026 version.
Instructions
- In the Training Data Manager, select a sample whose label needs to be modified.
- In the Label Editing group, select the desired tool and complete the editing operation:
- Positive Label: Draw a polygon on the image in the map to add a positive sample area.
- Negative Label: Draw a polygon on the image in the map that intersects with the existing positive label to subtract that area (mark it as negative sample/background).
- Edit Node: First click on an existing polygon in the map to select it, then use this function to add, delete, or drag nodes to adjust its shape.
- After editing, click the Save button. The system will update and generate a new label file based on the drawn positive and negative label areas. The new label file will be automatically saved to the Modified folder under the project storage path. The sample's group in the Training Data Manager will be automatically classified as Modified.
Related Topics