Electronic Arts Inc.
EXTENDING KNOWLEDGE DATA IN MACHINE VISION

Abstract:

A machine-vision system obtains a trained-synthetic dataset associated with a virtual-sporting event, features of the trained-synthetic dataset including features associated with the virtual sporting event. The machine-vision system can further train by extending the trained-synthetic dataset using a real-life dataset associated with an actual sporting event including recognized results, the extending can include identifying and selecting a portion of the recognized results from the real-life dataset for annotation, an unsupervised machine-learning component annotating the portion, adding the annotated portion to the trained-synthetic dataset to obtain an extended-trained dataset, based on improvement of detection of the features from the extended-trained dataset compared to the trained-synthetic dataset, and availability of another portion of the recognized results, repeating the extending using a part of the extended-trained dataset as the trained-synthetic dataset in a next iteration, and providing the extended-trained dataset to the machine-vision system.