Glossary of Useful Terminology

What we'll cover

Glossary of useful terminology used in Alegion Control.

• Assets: unlabeled data like videos you add to Control to be annotated. Once added, an asset can be used across multiple projects.
• Association:  is used to track relationships between two entities. For example, a product could be associated with a shopper or a wheel associated with a vehicle. Associations can persist for the duration of a video or only specific frames.
• Classification: is used to define a property of an entity. For example, an entity that represents a player in a basketball game could have a classification that indicates team membership or a temporal classification that indicates possession of the ball.
• Entity: is an instance of a class in an image or video. Each entity is given a label and localized with a shape such as a keypoint or bounding box. An entity may also have its properties classified and can have associations with other entities. For example, in a use case involving a basketball game video,  players as entities that are localized with bounding boxes. The players may be classified based on the team they belong to, and players could have a relationship with a ball entity to indicate ball possession.
• FFMPEG: an open source utility that can be used to get the frames per second for video files and diagnose and remediate a variety of issues with video files.
• Frames Per Second (FPS): every video file has a frames per second property that indicates it’s playback speed. Control uses this property to calculate the current frame location when annotating and it is critical that this property is specified accurately when adding a video as an asset.
• Interpolation: a feature of Control that automatically adjusts the position of an entity between any two position keyframes. Interpolation allows annotators to skip the annotation of frames where the software can automatically calculate the location with a required level of accuracy.
• Intersection Over Union (IOU): is a metric that relates the overlap (intersection) between two bounding boxes and the total area (union) taken up by both boxes. The intersection area is divided by the union area to get the IOU ratio.
• Json: JavaScript Object Notation format Control uses for annotation output.
• Keyframe: when annotating a video, a keyframe is a frame of the video on which the annotator marks a change in position, visibility, classification, or relationship. Keyframes record changes explicitly marked by the annotator, as opposed to changes that are calculated by the software automatically.
• Localization: Using a bounding box, keypoint, or other shape to mark the location of an entity in an image or video frame.
• Ontology: is the definition of the entities, classifications, and associations that determines how data is labeled. For example, a simple labeling project could have an ontology in which vehicles are defined as entities that are localized with a bounding box, there is a classification that specifies the type of vehicle, and an association between a vehicle and another vehicle such as a trailer.
• Pixel Distance: is measured by the straight line distance between two point in the pixel coordinate system of an image or video frame. Also called Euclidean Distance or Cartesian Distance.
• Project: is the records to be annotated, the ontology that defines how the annotations are structured, and the annotation results for completed records.
• Record: is created when an asset is added to a project. The record captures all of the annotation information for the asset in that project.
• Visibility: is a special classification that is automatically added to every entity that is annotated in a video project. This classification is used to indicate when an entity is “In Frame” or “Out of Frame”. When an entity is created, it is marked as In-Frame. When an entity exits the frame it can be marked Out of Frame. This records the state of the entity in the output data and hides the shape used to localize it. When the entity re-enters the frame it can be marked In-Frame again. This makes it simple to track objects that move in and out of frame while maintaining entity persistence throughout the entire video.