eBay had 28 patents in artificial intelligence during Q3 2023. These include search techniques that locate and display prospective objects in digital images based on a removed object, a categorization analysis system that identifies miscategorized listings, a machine that determines image quality scores and presents search results based on the scores, a videoconference system that generates a detailed view of the active speaker, and computer vision techniques that monitor user interaction with digital content to provide personalized recommendations and marketing content. GlobalData’s report on eBay gives a 360-degreee view of the company including its patenting strategy. Buy the report here.
eBay grant share with artificial intelligence as a theme is 50% in Q3 2023. Grant share is based on the ratio of number of grants to total number of patents.
Application: Prospective object search techniques based on removed objects (Patent ID: US20230260000A1)
The patent filed by eBay Inc. describes search techniques for locating and displaying prospective objects in digital images based on a removed object. The method involves obtaining a digital image depicting a physical environment and displaying it in a user interface. An object depicted in the image is then removed, and an aspect of the removed object is identified. The system leverages this aspect to locate a prospective object by filtering out candidate objects that have the same aspect. The prospective object is then configured for display within the digital image, and the image is displayed in the user interface with the configured prospective object depicted in the physical environment.
The identification of the aspect of the removed object can be based on a model trained by machine learning. This model can use data generated from objects depicted within the digital image, with the removed object serving as a negative sample and the remaining objects serving as positive samples. Alternatively, the aspect can be identified based on a generated vector representation of the removed object.
The method also includes determining the candidate objects from a database of listings of objects available for purchase. The removal of the object can be done in response to user input indicating the removal. The aspect of the removed object can be identified by determining that the remaining objects depicted in the digital image do not have the same aspect.
The patent also describes a computing device that performs the method, with a processing system and a computer-readable storage medium containing the necessary instructions. The device can display the digital image, remove the object, identify the aspect of the removed object, locate the prospective object, configure it for display, and display the image with the prospective object in the user interface.
Additionally, the patent covers computer-readable storage media that contain instructions for performing the method. These instructions, when executed by one or more processors, display the digital image, remove the object, identify the aspect of the removed object, locate the prospective object, configure it for display, and display the image with the prospective object in the user interface. The filtering of candidate objects can involve either filtering for objects that have the aspect or filtering out objects that have the aspect. The configuration of the prospective object can involve learning an aesthetic location for its display based on a model trained by machine learning, which includes factors such as relative size, orientation, and distance to the remaining objects in the image.