Create Autolabel jobs

Labeling jobs take the label spec and execute the labeling flow on the input images from the catalog. This article provides the steps to create a label job.

1. In the left navigation pane, navigate to Repo > Datasets.

2. Locate the required Dataset for which you want to create a label job, and click Catalog to open the catalog page.
   

3. Click the down arrow head adjacent to the Visualize button on the top-right corner of the catalog page, and select Auto Label.
   

4. In the New Labeling Job screen, enter the Job Name and Job Description.

5. Select Job Type as Labeling Classification or Labeling Object Detection, depending on the spec type for which you are creating a labeling job.

6. Select Docker Image.

   • AkridataAutoLabelObjectDetection: This is a higher accuracy labeling flow but is expected to take longer.

   • AkridataAutoLabelObjectDetectionFast: This option can be used to run a quick trial but will give a lower accuracy.
     

7. Select Labeling Spec.
   Based on the Job Type you have selected (Labeling Classification or Labeling Object Detection), the related label specs will appear in the list.
   You can also click Create Spec to open the Label Spec creation flow.

8. Specify the Confidence Threshold.
   By default, the value is 0.2. Auto labels below this confidence threshold will be filtered out from the results.
   

9. Set other options as applicable for the selected Job Type.

   • For Labeling Classification

     • Region of interest: Enable this option to mark an area in the image that should be considered for the job.

   • For Labeling Object Detection

     • Max Objects Per Image: Specify the maximum number of objects to be detected in an image.

     • Nestedness threshold: If two objects with Intersection over Union(IOU) of more than this threshold will be considered nested objects.

     • Region of Interest: Enable this option to mark an area in the image that should be considered for the job.

     • Show Segments: Enable this option to consider the object and draw the object's outline as a segment.

     • Remove same class nested objects: Enable this option to filter out nested objects of the same class.
       

       The region of interest is currently ignored for labelling and is used only to restrict the part of the image used to show the plot view.

10. Click Submit.
    A new labeling job gets submitted.

11. Navigate to Label > Jobs to view the job on the Labeling Jobs screen.
    The job submitted appears as Processing. Once the labeling process is complete, the job appears as Ready.

    1. Label Jobs created with Labeling Classification are listed with the label LA-CL.
       

    2. Label Jobs created with Labeling Object Detected are listed with the label LA-OD.