Use the Defect Segmentation Module

You are viewing an old version of the documentation. You can switch to the documentation of the latest version by clicking the top-right corner of the page.

Please click here to download an image dataset of network ports, an example provided by Mech-DLK. Different from the content in Train Your First Model, in this section, we will use a Defect Segmentation module and train a model to detect defects such as deformations and fractures on the network ports.

You can also use your own data. The usage process is overall the same, but the labeling part is different.

Workflow

  1. Create a new project and add the Defect Segmentation module: Click New Project in the interface, name the project, and select a directory to save the project. Click example projects icon create in the upper right corner of the Modules panel and add the Defect Segmentation module.

    example projects add new module
  2. Import the image data of network ports: Unzip the downloaded file. Click the Import/Export button in the upper left corner, select Import Folder, and import the image folder. The pins in the images can be deformed, fractured, or intact.

    example projects import images
    When you select Import Dataset, you can only import datasets in the DLKDB format (.dlkdb), which are datasets exported from Mech-DLK.
  3. Select an ROI: Click the ROI Tool button example projects icon roi and adjust the frame to select the pins in the image as an ROI, and click Apply to save the settings. Setting the ROI can avoid interferences from the background and reduce processing time. Setting the ROI can avoid interferences from the background.

    example projects roi
  4. Labeling: In this example, you will need to label the OK images and NG images in each dataset. OK means that the connectors meet quality requirements and NG means that there are defects such as deformations and fractures on the Ethernet ports.

    example projects ok defect features
    • For NG images, right-click the Polygon Tool and select a suitable tool to select the defect regions. In this example, it is recommended to use the Brush Tool. During labeling, the brush should be close to the edge of the defect to avoid the inclusion of a large number of non-defective areas. Click here to view how to use labeling tools.

      example projects label data
    • For OK images, select an image and click the OK Label OK label tool on the labeling toolbar. The image will be labeled as OK.

      example projects label ok
  5. Split the dataset into the training set and validation set: Please make sure that both the training set and validation set include images with all types of defects and at least one OK image. This can guarantee that the algorithm can learn all different types of defects and validate the images with different defects properly. If the default training set and validation set cannot meet this requirement, please right-click the individual image and switch it to the training/validation set manually.

  6. Train the model: Keep the default training parameter settings and click on Train to start training the model.

    example projects training chart
  7. Validate the Model: After the training is completed, click Validate to validate the model and check the results. You can also modify the Defect determination rule to filter results.

    After you validate a model, you can import new image data to the current module and use the pre-labeling feature to perform auto-labeling based on this model. For more information, see Pre-labeling.

    example projects result verification
  8. Export the model: Click Export and select a directory to save the trained model.

    example projects model files

The exported model can be used in Mech-Vision and Mech-DLK SDK. Click here to view the details.

We Value Your Privacy

We use cookies to provide you with the best possible experience on our website. By continuing to use the site, you acknowledge that you agree to the use of cookies. If you decline, a single cookie will be used to ensure you're not tracked or remembered when you visit this website.