GuidesRecipesAPI ReferenceChangelogDiscussions
Log In

Smart Labeling

Check how to use smart labeling.

What is Smart Labeling?

It is a service that before users proceed with the labeling task, provides the inference result of the AI model so that the users can efficiently proceed with the task.

A method to use smart labeling

1. Applying smart labeling

Select the data that you want to use smart labeling as a unit of folder or file and then click the Smart Labeling button in the upper right corner.

2. Setting up smart labeling inference

Smart Labeling setting screen

Smart Labeling setting screen

  1. Depending on the "labeling settings"-"task type" of the project, the types of models that can be selected change.
  • Bounding box = Object Detection Task Model
  • Poly segmentation = Instance Segmentation Task Model

  1. At Select task AI model, select the model that you want to use from the Preset AI list at the bottom.
  • If there is Custom Model learned from the same customers’ data, that Custom Model can be selected as well.

  1. Apply Smart Labeling by completing "AI Model select" , then proceed "Model Class Selection".
  • Set up which class you want to infer from the selected model from the class that is set up within the project.
  • It is possible to select only the class you want without setting all the classes.

  1. After finishing Select model class, proceed to "Set others".
  • When checking Instance size, based on the model inference results, instances that are smaller than the project’s minimum size should not be inferred.
  • Threshold standard means the confidence level (score) of the result inferred by the model and only Instances that the higher the threshold value, the higher the confidence level are available to be inferred. (When you set it to 0.7, objects which have a confidence level of less than 0.7 are not inferred.)
  • Preview inference results is a function that before applying Smart Labeling to all the selected data, it checks Smart Labeling performance according to the setting value as much as the "number of preview files".
  • After you check the performance by using Preview inference result, it is possible to change the setting value and apply it again.

  1. After completing all the settings, click the completion button at the bottom right.

3. After applying smart labeling

After applying smart labeling

After applying smart labeling


Example of applying Smart Labeling

  1. After applying Smart Labeling, an icon will be created next to the file name.
  • Green: Smart Labeling application is successfully completed
  • Gray: Although Smart Labeling is successfully applied, there is no inferred instance in the image
  • Red: Because of the Smart Labeling processing error, Smart Labeling has not been applied(When red comes out, you should apply Smart Labeling again.)

  1. As the AI model that is selected by the example of applying Smart Labeling image, the result inferred by the setting value will be printed out.

  1. Workers are using the Preset-Smart Labeling inference results so that they proceed with the inspection or labeling tasks again.

If you have any other inquiries, please get in touch with us at [email protected]