Key concepts
Below follows descriptions of common concepts used by Kognic, in the Annotation Platform and this documentation.
Each project comes with its own guideline. It will describe what you want to be annotated/the client wants you to annotate, how to annotate it, and give you visual examples of how to interpret the given instructions.
An input batch is a group of inputs. An input is a set of sensor data you want to annotate. It usually consists of at least one image/frame. It can also contain images from several cameras, a point cloud, a video, or a sequence of images.
Annotation work on different tasks or task types is always grouped by projects. The projects and their content is defined by Kognic and its clients.
Annotation Request is the major concept in the organization of annotation work at Kognic. A request can also be viewed as a subproject. There can be multiple types of requests in one project. The type of request determines what interface and annotation tool you/or the annotators will work in/ with.Β
A single request ties together input data with the annotation guideline, the taxonomy, the annotator/reviewer/supervisor team, and configuration of the annotation tool as well as the workflow. The request has all the information needed to produce the annotations on the input data.
On the input's way toward getting a delivery-ready annotation, it goes through a series of workflow stages. Each stage corresponds to one action in one of the workflow's different phases, such as the action βCorrectβ in the phase βReviewβ.
Which stages the input goes through depends on the workflow used in the request. An input can only be in one workflow stage at a time.
Tasks are created in each of the request's workflow stages and become available for team members to complete. The most common task actions are annotate, correct, and review.
Annotate An annotate task will be assigned with the purpose of being annotated from scratch.
Correct A task where you are asked to correct a task with existing annotations. The annotation you are asked to correct can be from someone else or your own.
In the Kognic platform, we have multiple different types of correction tasks, you can read more about them under Introductionο»Ώ.
Review In a review task, you are asked to determine if an annotation is of acceptable or unacceptable quality. If the task is of unacceptable quality you are also expected to provide correction requests and feedback to the annotator so they can improve the quality while becoming a more skilled annotator.
An object is a unique object in the world that can be annotated in multiple sensors or multiple frames. An object has a unique object id, which is visible in the annotation tool's object list.
A shape is the representation of an object in a specific sensor and in a specific time frame.
Examples:
- If you work in a sequence task with one sensor there can, for one object, exist a maximum of one shape per frame.
- If you work on an image task with one 2D sensor and one 3D sensor there can exist two shapes per object, one in 2D and one in 3D.
- If you work in a sequence task with one 2D sensor and one 3D sensor there can maximum exist two shapes per frame per object.
ο»Ώ
ο»Ώ
