Workflow


Developer

  1. New Project → Image Generation → Create a Developer project.
  2. Register the product to train in the Products panel on the [Database] tab.
    • Multiple products can be registered.
  3. Load images to train for each registered product.
    • Both OK and defect images are required for each product.
    • At least one defect label and an OK image are required for each product.
  4. Click the Edit Class [ tool_edit ] button in the Classes panel of the [Database] tab to add the defect class to generate.
  5. Label the images of each product.
    • Label the OK and defect images in the same way as the Segmentation project.
  6. After labeling is complete, go to the [Preprocess]-[Align] tab and set a reference image (ref. Image) for each product and then align using the Alignment algorithm suitable for the image.
  7. After aligning, go to the [Preprocess]-[Set Region] tab and mark the region where each defect class occurs on the image of each product.
  8. Go to [Train] and set the training information.
  9. Enter the name and description of the set training and click the [Add List] button.
  10. When all training lists are added, click the [Train] button.
  11. When the Training Progress window pops up, click the Play button on the Training List panel to start training.
  12. When the training is completed, click the [Finish] button.

Generator

Generator generates defect images using the training result from the Developer. Generator has Auto mode and Manual mode.

Auto mode

  1. New Project → Image Generation → Create the Generator project.
  2. Select the Auto mode.
  3. Click the [ tool_add_tab ] button to select the model file trained in the Developer, and then select whether to generate defects on the product used for training or on a new product.
  4. When using the product used for training in the [Align] tab, use the reference image set during training. When selecting a new product, set the reference image after loading the image to generate a defect.
  5. Load the image (OK image) to generate a defect in the Template Image panel.
  6. Align using the Alignment algorithm suitable for the image.
  7. After aligning, set the region to generate defects for each defect class in the [Set Region] tab. When generating a defect on a product used in training, users can change the region information set in training to a desired region.
  8. After the region is set, clicking the Generate button in the lower right of the [Generate] tab displays a pop-up window. In the pop-up window, set the number of images to generate defects, the maximum number of defects to generate per each image, whether to generate gray scale patches, the percentage values to generate for each defect (Weight), and the minimum size (Size Threshold) and then click the Generate button to generate the image.
  9. Check the status of the defects that have been generated, and then delete the defects of poor quality.
  10. After the defect is generated, click the Export button on the bottom right to export the image where the defect was generated in .srImage format and export the SaigeVision Segmentation project (.srproj) together where necessary.

Manual mode

  1. New Project → Image Generation → Create the Generator project.
  2. Select the Manual mode.
  3. Click the [ tool_add_tab ] button to select the model file trained in the Developer.
  4. Load the image (OK image) to generate the defect in the Template Image panel.
  5. Draw the shape to be generated for each defect on the loaded image.
  6. After drawing the shape of the defect to generate, click the Generate button at the bottom right to display a popup window. In the pop-up window, set the maximum number of defects to generate for each image, whether to generate gray scale patches, and the image to generate, and then click to Generate button to generate the image.
  7. Check the status of the defects that have been generated, and then delete the defects of poor quality.
  8. After the defect is generated, click the Export button on the bottom right to export the image where the defect was generated in .srImage format and export the SaigeVision Segmentation project (.srproj) together where necessary.