To create a new task, navigate to the tasks tab within a project and click ‘new task.’ The popup will guide you through task creation, and the documentation below will provide details on each step. Once you create a task, the task type and consensus value cannot be changed, but all other properties may be edited.
Classification tasks allow the annotator to check boxes presented with an image to indicate the presence of one or more classes. Although classification tasks use the canvas to display images and associated questions, no image annotations are made in these tasks.
Object Detection Task
Object detection tasks allow the annotator to place bounding boxes around each instance of a class that’s present in an image.
Semantic Segmentation Task
Semantic segmentation tasks allow users to identify all members of a certain class with pixel-level accuracy. Semantic segmentation tasks do not allow the annotator to distinguish between different instances of the same class. Because segmentation tasks require pixel-level accuracy, Innotescus provides a much more sophisticated set of tools than for object detection tasks.
Instance Segmentation Task
Instance segmentation tasks are just like the semantic segmentation tasks described above, only they do require the annotator to distinguish between different instances of the same class. Annotators will still have access to all the same tools they have in semantic segmentation tasks to ensure pixel-level accuracy.
The task creation process allows users to apply tasks to one or more datasets within the same project. Users can add a new dataset at this stage if need be, apply to an existing dataset, or apply tasks to datasets later on from that project’s datasets tab. To assign a task to a dataset after task creation, see this section.
Task creators may add classes by typing each name individually and pressing enter, pasting a list of comma separate classes, or by using the 'import classes' button to import a .csv of class names or copy classes from an existing task. Class names can be up to 256 characters long.
If importing classes using a .csv, the .csv file must only include a comma-separated list of class names, like the sample below:
Each new class is assigned a color once it is entered, but you can edit a class' color by clicking the colored circle next to its name and choosing a new one; classes that already exist within the project will share a color, so editing a class color in one task will change it in the other.
Adding Classes from the Canvas
Using the check box available below the class entry line, task creators can allow annotators to add new classes directly from the annotation canvas.
Innotescus’ consensus algorithm automatically merges annotators’ work to ensure higher quality annotations for your dataset. Select the number of annotators you’d like to have annotate each image before the annotations are merged into the consensus output. Consensus is available on all tasks except for classification tasks.
Creating Tasks with Imported Annotations or Pre-annotations
Users can also create new tasks through the annotation import process. This method of task creation uses the import file(s) to create classes, and currently defaults to a consensus of 1.
Annotations can be imported as complete annotations or as pre-annotations, depending on the supervisor's preference. Pre-annotations are presented with each image during the annotation process, and are a great way to improve annotator efficiency.