Introduction to Labeling Tools
OK Label
Use the OK Label to label an image as OK.
-
Select an image from the image list and then click the OK Label on the toolbar. The image will be labeled as OK.
Defect Labeling Tools
Defect labeling tools can label the defects in images to provide the information required by deep learning training.
You can choose among the following labeling tools built in the software according to actual needs.
Polygon Tool
The Polygon Tool can draw polygon labels with more vertices, which is suitable for regular defects.
-
Click (or press P on the keyboard).
-
Click the first position (vertex) in the selection area, then click the second one, third one, etc., to draw the labels, and right-click to finish.
Brush Tool
Use the Brush Tool to draw labels of any shape. This tool is suitable for defects with complex shapes.
-
Right-click and then click (or press B on the keyboard).
-
Adjust the slider to set the thickness of lines according to the size of defects.
-
Long press the left mouse button in the selection area, move in any direction, and then release the left mouse button to finish the drawing.
Autofill Lasso Tool
Draw arbitrarily shaped labels by forming closed shapes with brush paths. This tool is used for defects with complex shapes.
-
Right-click and then click (or press A on the keyboard).
-
Adjust the slider to set the thickness of lines according to the size of defects.
-
Long press the left mouse button in the selection area and move in any direction to form closed shapes with the starting point and thus finish the drawing.
Auxiliary Labeling Tool
After model training, use the Auxiliary Labeling Tool to label images without labels.
Right-click and then click (or press Shift + A on the keyboard) to complete auxiliary labeling.
Smart Labeling Tool
The Smart Labeling Tool can be used to automatically select the objects in the image.
When multiple objects in an image have large color differences and are scattered, you can use the Smart Labeling Tool to conveniently label the objects in the image.
-
Click (or press M on the keyboard).
-
Move the cursor in the selection region and then click the object to be labeled.
-
If the selection cannot completely cover the object, click the uncovered part to expand the selection area.
-
If the selection covers the areas outside the object, right-click these areas to reduce the selection area.
-
-
Click Apply in the upper-left corner of the selection region.
You can use the Select Tool to fine-tune the labeled contour by the following steps:
-
Use the Select Tool to select the label to be adjusted.
-
Adjust the contour in one of the following three ways according to actual situation. Please ensure that the selected area closely aligns with the object contour.
-
Place the mouse cursor on a vertex of the contour. When the cursor turns into , long-press the left mouse button and drag the vertex to adjust the contour.
-
Place the mouse cursor on a vertex of the contour. When the cursor turns into , click the right mouse button to delete the vertex.
-
Place the mouse cursor on the contour. When the cursor turns into , click the left mouse button to add a vertex.
-
When the objects differ greatly in colors and have clear contours, it is recommended that you label multiple objects at a time and click Apply. If the objects are not obviously distinct, it is recommended to label one at a time. |
Currently, the Smart Labeling Tool can work in two modes, and it will automatically switch between them according to the actual situation. Users do not need to manually select a mode.
-
Standard mode (GPU): In this mode, the Smart Labeling Tool has robust labeling capabilities. This tool will automatically switch to the standard mode if the running device has a GPU, with available GPU memory of at least 2.5 GB, and there is no project undergoing training or validation.
-
Lite mode (CPU): In this mode, the Smart Labeling Tool has basic labeling capabilities. This tool will automatically switch to the lite mode if the running device has no GPU, or the available GPU memory is less than 2.5 GB, or there is a project undergoing training or validation.
Mask Tool
If there are some irrelevant parts that may interfere with model training/inference, you can use the Mask Tool to cover such parts. The masked parts will not be involved in training/inference. For example, the object surface features that should not be judged as defects but are similar to defects, need to be masked out.
You can choose among the following three mask tools built in the software according to actual needs.
|
Mask Polygon Tool
-
Click (or press Shift + P on the keyboard)
-
Set Mask Type and Mask fill.
-
Click the first position (vertex) in the selection area, then click the second one, third one, etc., to draw the labels, and right-click to finish.
Labeling Eraser Tool
The Labeling Eraser Tool can be used to erase the labeled region.
-
Click (or press E on the keyboard)
-
Long press the left mouse button in the selection area and move in any direction.
Adjust the slider to change the eraser size.
Mask Eraser Tool
The Mask Eraser Tool can be used to erase the masks.
-
Right-click and then click (or press Shift + E on the keyboard)
-
Long press the left mouse button in the selection area and move in any direction.
Adjust the slider to change the eraser size.
Grid Cutting Tool
In industrial inspection scenarios, if the size of images captured by the camera is large, smaller defects may be inconspicuous. If training in such a case is performed, defects are difficult to detect. You can use the Grid Cutting Tool to cut the large images into cell images of the same size according to the set dimension. Defect labeling should be completed for all images before the application of this tool.
-
Click (or press U on the keyboard).
-
Set Rows and Columns and then click Apply.
-
Place the cursor in the parameter boxes and then scroll the mouse wheel.
-
Enter values in the parameter boxes.
Note that the number of rows and columns should not be too large, or else the number of cell images after cutting is great, which slows down subsequent inference.
-
Grid Selection Tool
Right-click and then click (or press I on the keyboard) to open the Grid Selection Tool. By default, the cell images with defect labels are checked, which will be added into the training/validation set. You can select cell images with and without defects on demand. Click the Preview button in the upper right corner of the selected image to preview the cell images.
-
Select defects: Select all the cell images containing defects to put into the training/validation set.
-
Select all: Select all the produced cell images into the training/validation set, and set those containing no defects to OK.
-
Clear selection: Clear the selections on cell images.