Create Annotation Task
Last updated
Last updated
On the launch page of the annotation task, different roles in DataVerse have different levels of authority to view the list of annotation tasks. While Owners, Admins, and Members can view all the annotation tasks in this workspace, Specialists can only view the tasks they are assigned to execute.
This includes:
Name: The task name.
Data Slices: Select a data slice from the drop-down menu. In the current version, "image", "frame" and "sequence" are supported.
Pre-Annotation: Any existing annotations is this project can be imported as a pre-annotation to boost the labeling. If you don't need pre-annotations, simply select "None".
It is fine that a data row not having selected pre-annotation; it will simply shows no pre-annotation objects in the labeling page.
In the first step of creating annotation task, users can further make the following settings in a specific task.
When the ontology type is '2D bounding box,' users have the option to configure whether the bounding box can extend beyond the boundaries of the image. The default setting is NOT allowing bounding box extends beyond the boundaries of the image.
When the data slice is "sequence type", users can set the annotation sampling larger than 1 so that not each image in each sequence will be loaded in to this task for labeling. The default setting is 1, which means each image in each sequence will be loaded for labeling.
Both advanced settings are not changeable after a task is created.
In principle, the task ontology inherits the project ontology but some modification could be made to facilitate the labeling task. The modification includes:
The class to label in this task: This means that it is NOT necessary to label all the classes in this project. If you have many classes, splitting them into two tasks to label different classes may increase efficiency in the labeling task.
Set the attributes of each class to be required or optional.
If the attribute is option question, specify whether it is single-answer or multiple-answer.
Please be noted that the task ontology cannot be further modified once the task is created.
If you apply "default setting", the the task ontology will be:
All the classes can be labeled
All the attributes are optional to answer
All the option questions of the attributes is single-answer.
You may also input the instructions for each class, and the instructions can be further edited after the annotation task is created.
Similar to the task ontology, some modification can be made to the questions of the tag. The modification includes:
Whether the tag should be labeled, and which items in the tag should be labeled.
Whether those to-be-labeled questions is optional or required to answer.
If it is an option question, specify whether is is single-answer or multiple-answer.
Please be noted that the tag setting cannot be further modified once the task is created.
If you apply "default setting", the the task ontology will be:
All the questions in the tag can be labeled.
All the questions are optional to answer.
All the option questions is single-answer.
While creating an annotation task, it may take some time to import the required information. Once the task is ready, the status will be “Ready", and the task name is clickable to access the the Task Overview.
While the status of a task is "Processing", the Task Overview is unaccessible.
You must select a specific project prior to creating an annotation task. Clicking and fill the information of an annotation task.
Similarly, you can also input the labeling instruction for the tag, and the instructions can be further edited after the annotation task is created.