Users can only annotate datasets by working on the tasks assigned to them, or by uploading corresponding annotations.
Annotating Specific Images
Users can annotate a specific image either by selecting 'annotate' from that image's options menu ( ⠇) and selecting the task to annotate, or by viewing that image in the image viewer, and clicking the 'annotate' button there. If multiple tasks are assigned to that image, the user must choose the task from the options that appear before annotating. If another user is annotating or assigned that image/task combination, a message will pop up saying so.
Editing Dataset Annotations
Users can edit existing annotations by clicking the ‘edit’ or ‘edit consensus’ button that appears in the image or video viewer. For consensus annotations, users can edit an intermediate annotation set by selecting it before clicking the 'edit' button.
Editing an annotation set brings a user back to the annotation canvas, where users can change and submit annotations just like any other task.
To import annotations for an existing dataset, click on the options icon ( ⠇) for the dataset in the datasets tab and select ‘import annotations.’ Annotations can be imported in the coco format or as a collection of masks with an associated classes.txt file. See this page for details on import formatting.
Creating a New Task
When creating a new task, users must select the format of the annotations being imported and the task type of the task being created using the available dropdowns.
Importing to an Existing Task
When importing annotations to an existing task, users must first make sure the task is assigned to the dataset for which annotations are being uploaded. Then, users can proceed with the normal annotation import process, and can now optionally overwrite any existing annotations that conflict with the annotations being imported using the available checkbox.
Pre-annotation and Overwriting Annotations
Supervisors can choose for the imported annotations to overwrite existing annotations or to act as pre-annotations. Pre-annotations are served up with their respective images or videos during the annotation process, and are a great way to correct model-generated annotations and improve annotator efficiency.
To assign tasks to a dataset, click on the options icon ( ⠇) for that dataset and select ‘Assign Tasks.’ In the popup that appears, you will be able to assign the remaining tasks to that dataset.