iTAG is an intelligent data labeling platform of Platform for AI (PAI). iTAG allows you to label different types of data, such as images, text, videos, and audio. iTAG also allows you to label multimodal data. iTAG provides a variety of labeling content and topic components. You can use common labeling templates that are provided by iTAG or create custom labeling templates based on your business scenarios.
Procedure
Perform the following steps to label data in iTAG:
In the dataset management module, create a dataset for the data that you want to label. A .manifest index file is generated.
Use a common or custom labeling template in iTAG to create a labeling job based on the dataset that you created, and then distribute job packages. You can complete a labeling job in the following phases: labeling data in job packages, reviewing labeling results, and accepting the job packages. The first phase is required. The last two phases are optional. When you create a labeling job, you can specify one of the following combinations of phases for the job: labeling phase, labeling and review phases, labeling and acceptance phases, and labeling, review, and acceptance phases. The following descriptions provide the operations in each phase:
Labeling: On the Label Task page in the iTAG console, a labeling worker claims a job package. Then, the labeling worker labels the data in the job package and submits the job package.
Review: On the Quality Inspection Task page in the iTAG console, a labeling worker claims a job package whose data is labeled. Then, the labeling worker can review, modify, or reject the labeling results.
Acceptance: On the Acceptance Task page in the iTAG console, the person who requires the labeling results claims a job package, reviews the labeling results in the job package, and then accepts or rejects the job package.
Complete the labeling job by following the specified phases and obtain the labeled data.
Export the labeling results to the Object Storage Service (OSS) bucket that you specify. Then, you can use the results to train models.
Data formats
Format of input data
When you create a labeling job, use a dataset in the .manifest format as the job input dataset. You can create a dataset in the dataset management module of PAI. Then, PAI automatically generates a .manifest file that is required by labeling jobs. For more information, see Create a dataset for a labeling job.
Format of output data
In iTAG, you can export labeling results to a .manifest file. For more information about data formats in the labeling, review, and acceptance phases, see Export labeling results.
Contact us
If you have questions or issues related to iTAG, join the DingTalk group 21930006619 for technical support.