An annotation profile defines the behaviour of the annotation interface. You will need to create a profile before you can do any annotations in SAVANT. Profiles also live at the workspace level, so you can re-use profiles across projects if you need to.

SAVANT has a centralised registry for labels that work for the entire workspace. You can find all the labels in the “Labels” section of the left-hand navigation. In here, you can consult all the labels that are available in your workspace and you can create new labels by clicking on the “Create label” button on the top right hand of the screen. To create a label you will need to provide:

Once you have created all the labels that you need for your annotation project, you are ready to create an annotation profile.

You will see that there is an option to create “skeletons”. These can be used to create groups of tags that should all be annotated together. These will be covered in a future guide.

You can find all the annotation profiles that have been created in the workspace in the “Profiles” tab of the left-hand navigation. From there, you can create a new profile by clicking on the “New Profile” button located at the top right of your screen. You can also consult existing profiles by clicking into any of them. Before filling-in any other fields, you will need to provide a unique name for your profile, a description and upload a written version of the guidelines that you will use for annotations.

The written guidelines should be a PDF document that explains how the annotations will be done. This is typically required by quality-compliant annotations and it should help your annotators resolve doubts about how the annotation job should be done. You should also use this document to verify that the profile that has been configured in SAVANT matches how you expect the annotation to be done.

To enforce traceability of the annotation process, the profiles are immutable once they are created; to update the profiles, you will create new versions of them and all versions are accessible to you.

In the annotation/phase definitions section of the profile you can configure the labels that will be available for annotators when doing the annotation job. Add any labels from the label registry that you would like to enable annotators to use for the job.

Once you have added a label, you can configure it with the following actions:

It should be noted that the profile itself doesn’t force an annotation to be either timestamp or frame; this will be controlled later by the sequence type. The profile only controls what will be made available when either the video or frame annotation interfaces are used (see following guides on how to do annotations in SAVANT). As such, a user might create a single annotation definition (e.g. for an organ) and the same profile can be used to do both timestamp and frame annotations.

In addition to the primary video and frame annotations and tags, SAVANT allows other types of labels. These are generally called “tags”, and they can be thought of as additional metadata. SAVANT enables the following tags:

Within the profile, you can also specify automated checks to be performed on the annotations. For every check, you can decide whether to treat it as a warning (user is alerted of the rule violation) or as an error (user is prevented from completing the annotation). SAVANT offers the following QA rules.

Within the frame annotation interface, and for a given label, you can define a bounding box. Then, the rule will check that all the keypoints annotated for that label are within the previously defined box. Only labels which have bounding box and keypoint shapes can be added to this rule.

Within the frame annotation interface, it enables to create a bounding box with one label and force other labels to be defined within that box. This rule is similar to keypoint in bounding box, but can be used when the bounding box and the shapes are annotated with different labels.

Within the frame annotation interface, it defines a maximum percentage of the frame that a given annotation can cover.

It enables specifying a subset of labels that must be used within the sequence of frames that are being annotated. Other labels may also be used, the rule only guarantees that at least the labels selected are used.

Defines multiple subsets of labels that can only be used exclusively. That is, if a label from one subset is used, then only other labels from that same subset may be used. You can choose whether this rule applies per frame or in the entire sequence of frames.

Within the frame annotation interface, it defines a square position within which the label should be annotated. This QA rule is similar to label in bounding box, but it defines the same region for every frame that is being annotated.

Within the frame annotation interface, it defines for any two labels whether one should always be smaller or bigger than the other. The rule checks that "relative label smaller/larger than label" is true. In the example below, assuming that the blue label is the relative label; then the rule applied is relative label (blue) smaller than selected label (red).

Within the frame annotation interface, it defines for any two labels where they must be located with respect to each other. You can specify the rule in three different ways:

Specify the position of the centre point of the relative label in relation to a selected label.

Specifies that there must be overlap between some points of the relative label and the selected label in a specific direction. The image below specifies that the inner boundary of the relative label (blue) must be to the right of the selected label (red).

Specifies that there must not be overlap between the points of the relative label and the selected label in a specific direction. The image below specifies that the outer boundary of the relative label (blue) must be to the right of the selected label (red)

Within the video annotation interface, it defines the minimum and maximum length for all time intervals annotated. The image below shows an example of applying the maximum phase length (blue) to two different phase annotations.

Within the video annotation interface, it defines for a given label the minimum and maximum number of times that this label can be used. The image below shows an example of applying a maximum count QA rule of three in a video sequence.

Within the video annotation interface, it defines the order in which time intervals must be annotated. The image below shows an example of phase order rule where the red phase is forced to be before the blue phase.

The final section of the profile allows you to control whether AI assistance is available to the annotators doing the annotation job; you can also decide which model to use for each type of task. SAVANT offers models for three assistance tasks:

Now that you have created an annotation profile, you are ready to create and manage an annotation project.