Mech-DLK 2.6.0 Release Notes
This section introduces the new features and improvements of Mech-DLK 2.6.0.
New Features
Added the Parameter Settings Feature for the Smart Labeling Suite
Mech-DLK provides the smart labeling suite for you to efficiently label datasets.
The smart labeling suite includes the following labeling tools:
-
Smart Labeling Tool
-
Pre-trained Labeling Tool (formerly “Pre-labeling Tool”)
-
Visual Foundation Model (VFM) Labeling Tool (formerly “Super-labeling Tool”)
Mech-DLK 2.6.0 has added the feature that allows you to configure parameters for the smart labeling suite. When you use the smart labeling suite, you can configure the following parameters:
-
In the menu bar, click , and enable Always load smart labeling model. This reduces the frequency of model unloading in open projects, minimizing the number of times you need to wait for the labeling model to reload.
-
In the Instance Segmentation and Object Detection modules, when you are using the smart labeling suite, you can click the Settings button in the upper left corner of the selection region to configure labeling parameters. For more information about parameter description, see Introduction to Instance Segmentation Labeling Tools and Introduction to Object Detection Labeling Tools.
Improvements
Optimized the Training Parameters for Defect Segmentation
Mech-DLK 2.6.0 has optimized the training parameters for the Defect Segmentation module. You can now train a High-accuracy or High-speed model.
Optimized the Algorithm for Fast Positioning
-
Optimized training process: The Set Template button and Quick Template Tool are removed. The image adjustment operation is now performed during model validation, and it affects only the validation results. For more information, see Use the Fast Positioning Module.
-
New labeling tools: Polygon Tool, Ellipse Tool, Rectangle Tool, Smart Labeling Tool, Mask Polygon Tool, Mask Brush Tool, Mask Lasso Tool, Mask Eraser Tool, and ROI Tool.
Optimized the Algorithm for Unsupervised Segmentation
Compared to the original algorithm, the optimized Unsupervised Segmentation algorithm increases training speed by 7 times, increases GPU inference speed by 2 times, and increases CPU inference speed by 10 times.
Supported VFM Labeling Tool for Text Detection
Mech-DLK 2.6.0 has supported the use of the VFM Labeling Tool in the Text Detection module. You can use this tool to label text areas in batches.
Optimized the Export Options for Cascaded Models
In Mech-DLK 2.6.0, you can export a single model or all models from the cascaded models.
Optimized the Add Module Window
Mech-DLK 2.6.0 has optimized the Add Module window, dividing the algorithm modules into Categorization, Locating, Inspection, and OCR algorithms. Additionally, functional descriptions for each module have been added to help you select the module that best fits your business needs.