Dynamic Classifier Selection Based On Class Skew

A classification system comprising:

• a general classifier configured to classify multiple aspects of content of an image stream;
• a template store including multiple different specialized classifier templates, each of the specialized classifier templates being configured to classify a subset of the multiple aspects;
• and
• a classifier selection system configured to re-train one of the multiple different specialized classifier templates to generate a specialized classifier for a particular subset of the multiple aspects, and to determine when to switch between using the general classifier to classify the multiple aspects of content of the image stream and using the specialized classifier to classify the multiple aspects of content of the image stream.

A computing device comprising:

• one or more processors;
• and
• a computer-readable storage medium having stored thereon multiple instructions that, responsive to execution by the one or more processors, cause the one or more processors to perform acts comprising:
• obtaining an image stream;
• determining when a specialized classifier can be used to classify a particular subset of multiple aspects of content of the image stream, the specialized classifier exploiting class skew of the particular subset of the multiple aspects, the determining including identifying when the particular subset of the multiple aspects dominates the image stream;
• accessing the specialized classifier to classify the particular subset of the multiple aspects;
• and
• using the specialized classifier rather than a general classifier to classify the particular subset of the multiple aspects in subsequent images of the image stream.