Skip to content

Auto-labeling in Custom AI

1. What is Auto-labeling in Custom AI?

Custom AI is a function that proceeds with manual labeling of objects that you want to label and auto-labels with artificial intelligence models created based on them.

이미지이름

2. Create Custom AI project

1) Upload Data

이미지이름

1-1) Click the Start labeling button under "LABELING AI | Labeling" or go to "DS2 DATASET | Dataset".

1-2) Select the data set in "ZIP" format to auto-labeled and click the Start Labeling button. At this point, the upload status of the data must be "Complete" before the project can be generated normally.

- Please check Upload Data

2) Create Project

이미지이름
1-3) Select the project name, description, and data category as "object recognition," then click Next.

1-4) You can see that the project you created has been added to the project list under "LABELING AI | Labeling".

3) Manual Labeling

이미지이름

3-1) At least 10 manual labeling for each object class you want to label is required before proceeding with auto labeling through CUSTOM AI.

3-2) Please check Manual Labeling Guide for additional classes and manual labeling.

3-3) Select 'Create CUSTOM AI' when manual labeling is complete.

4) Create CUSTOM AI

이미지이름

4-1) Utilize a small amount of manual labeling data to generate CUSTOM AI for customized labeling of my data.

4-2) Select the labeling type and the class to use for CUSTOM AI development and click Start AI Development.

4-3) You can check the start and completion of custom AI project learning by notification tab and email.

For more information about labeling types, please refer to Manual Labeling Guide.

You can check the start and completion of custom AI project learning by notification tab and email.

5) Select the best CUTOM AI

It is a feature that selects the best model generated through custom AI development and utilizes it for auto-labeling.

이미지이름
- Once auto-labeling is complete, you can click Select Best Custom AI to select the best custom AI model created.

5-1) Select the best auto-labeling model out of the four images five times in total to select the model for future auto-labeling. Duplicate selection is also possible.

5-2) Click the icon in the upper right corner of each image to view the image zoom.

5-3) Finally, when you click Finish Selection, the model is set, and the model set for auto-labeling is used later.

6) Auto labeling

이미지이름

6-1) If you select BEST CUSTOM AI after developing CUSTOM AI, you can start auto-labeling. Click Start Autolabeling on the dashboard.

6-2) Select Custom AI from "Select artificial intelligence type" and select the model you want to proceed with auto labeling.

6-3) When you select the preprocessing option Face Nonidentification, it mosaic all faces in the image.

6-4) Select the number of autolabelings, then click Start Autolabeling.

By pop-up window, AI development will begin normally. You can check the progress on the dashboard, and we will notify you by e-mail and notification when it is completed.

2-5. Advancement of inspection and labeling AI

이미지이름

1) You can verify that auto-labeled data is properly labeled, and correct mislabeled data.

2) Also, by regenerating CUSTOM AI with modified label data, you can upgrade it to a more accurate autolabeling model.