Platform For AI:Create a labeling job

Limitations

Only administrators and annotation administrators can perform labeling operations. To request the required permissions, see Manage workspace members.

Procedure

1. Go to the iTAG page.

   1. Log on to the PAI console.

   2. In the left-side navigation pane, click Workspaces. On the Workspaces page, click the name of the workspace that you want to manage.

   3. In the left-side navigation pane, choose Data Preparation > iTAG.

2. On the iTAG page, on the Jobs tab, click Create Task.

3. On the Create Labeling Job page, configure the following key parameters. Configure other parameters as needed.

   Parameter	Description
   Task Name	1 to 100 characters. Must start with a letter, digit, or Chinese character. Can contain underscores (_) and hyphens (-).
   Input Dataset	Select the dataset created in PAI.
   Template type	Supported template types: General Template: Preset template provided by the platform. Custom Template: Create a custom template by combining Content Components and Topic Components. For more information about custom template data formats, see Custom templates.
   Template	When Template Type is set to General Template, select one of the following templates: Image For use cases and data formats, see Image. OCR: Recognizes text within a selected area of an image. Object Detection: Detects and locates specific objects in an image. Image Classification: Classifies an image based on preset labels. PDF: Performs OCR and label classification on PDF files. Moderation and Matting: Performs moderation and matting on images. Table Recognition: Uses algorithms to pre-recognize key elements in a table. You can edit the results as needed. Text For use cases and data formats, see Text. Entity Recognition: Establishes relationships between text entities. Text Classification: Classifies text based on preset labels. This template supports single-label and multi-label classification. Entity relationship: Defines relationships between text entities for knowledge graph scenarios. Video For use cases and data formats, see Video. Video Classification: Classifies videos based on preset labels. This template supports single-label and multi-label classification. Audio For use cases and data formats, see Audio. Audio Classification: Classifies audio based on preset labels. This template supports single-label and multi-label classification. Audio segmentation: Segments the content of an audio dataset and adds labels to the segments. Audio Recognition: Transcribes text from audio content.
   OCR recognition results configuration	Available only when Template is set to Image > OCR. By default, OCR Recognition Results is selected. OCR recognizes text within a selected area of an image.
   Label Configuration	Enter label names for the job. Press Enter after each label name. For example, to identify cats in images, you can add labels such as "Cat", "American Shorthair", and "British Shorthair". This helps labelers apply labels to samples. Specify whether a single selection supports one or multiple labels. For single-label selection, set this option to Single Choice. For multi-label selection, set this option to Multiple Choice. For example, when identifying cats in images, if the label selection mode is set to Multiple Choice, the labeler can select a cat in a sample and apply both the "Cat" and "American Shorthair" labels. Note The Single Choice and Multiple Choice options determine the number of labels that can be applied to a single selection within a sample, not the number of selections that can be made on the sample.
   Enable Intelligent Labeling	For configuration details, see Data pre-labeling: Intelligent labeling configurations.
   Task Description	A brief description of the labeling job to distinguish it from other jobs.
   Assign Subtask Packages	iTAG groups labeling tasks into subtask packages that team members can claim and work on. Supported allocation rules: Fixed size: Each subtask package contains a fixed number of labeling tasks. The allowed package size range depends on the dataset size: If a dataset contains 0 to 20,000 items, the size of a subtask package can range from 1 to 200. If a dataset contains 20,000 to 100,000 items, the size of a subtask package can range from 5 to 200. If a dataset contains 100,000 to 500,000 items, the size of a subtask package can range from 25 to 200. If a dataset contains 500,000 to 1,000,000 items, the size of a subtask package can range from 50 to 200. Press Import Fields: Groups data into subtask packages based on a selected field in the dataset. All items with the same field value are placed in the same package. Targeted Assignment: Assigns subtask packages to specific labelers or teams.
   Check Proportion	If the Task Workflow you select includes a check step, such as Marking-Check or Marking-Inspection-Acceptance, you must configure the sampling ratio for the check process. The default value is 100%.
   User Configuration	Depending on the selected Task Workflow, assign the appropriate members or groups to roles such as labeler, checker, accepter, or task administrator. You can collaborate with multiple members of the workspace to complete the labeling job. For more information about role permissions in iTAG, see iTAG overview.

4. After configuring the parameters, click Create.

View and manage labeling jobs