Visual detection models localize an object of interest in an image by returning a box that bounds that object, as well as the type of that object, also referred to as the class. A detector can detect multiple objects of different classes per image. For each detection, a detector outputs a confidence score that is independent of other detections.
For Hive's logo detection model, the output object lists each logo detected in the visual input, including:
- A geometric description (coordinates) of a bounding box for each detection.
- The predicted class for the detection (typically a brand or company represented by the logo)
- A clarity score for each detected logo that represents visible identifiability – logos that are out of focus, not fully visible, or otherwise occluded will have lower clarity scores
- A location or display surface for the detected logo (e.g., a ball or stadium sign)
For a video input, Hive’s backend splits the video into frames, runs the model on each frame, then recombines the results into a combined response object. The video output for a detector is similar to a list of detection output objects, but with multiple timestamps for each sampled frame.
Hive’s logo model is the world’s most comprehensive logo model, boasting tens of millions of training labels and capable of recognizing 5,000+ logos. Our API's return detailed information on each logo in an image or video (location/size/clarity) as well as what kind of object the logo is placed on. Please contact our sales team for the complete list of logos or to inquire whether a specific logo is covered. Additional logos and locations can be requested on a rolling basis. For more information on use-cases and to see our models in action, see here.
Hive's logo model can detect logos on a wide variety of objects and locations, including:
- golf clubs
- basketball stanchions
- fan areas
- lower-level banners
- playing areas
- upper-level banners
- aerial stadium
- media backdrop
- stadium front
- home plate sign
- static sign
Updated 15 days ago