Back to results
US 12,755,522 B2Granted
System and method for classifying novel class objects
AssigneeElectronics and Telecommunications Research Institute
InventorsYe-Bin Moon, Yongjin Kwon, Jin Young Moon, Tae-Hyun Oh
Life of this patent
Estimating the term…
Abstract
Disclosed herein are a system and method for classifying novel class objects. The method of classifying novel class objects includes (a) constructing a novel classifier considering prior knowledge acquired from a base classifier, and (b) learning a parameterized weight coefficient of a novel classifier model during the training of the novel classifier.
Claims16
A method of classifying novel class objects, comprising:
- (a) constructing a novel classifier considering prior knowledge acquired from base classifiers previously trained for a set of base classes;
- and
- (b) learning parameterized weight coefficients of a novel classifier model during a training of the novel classifier,
- wherein parameters of the novel classifier model are generated as a linear combination of parameters of the base classifiers and a preset number of Gaussian random vectors used as additional basis vectors,
- the method further comprising:
- receiving an input image;
- processing the input image through a backbone network to obtain a first feature map;
- and performing object recognition for the novel class using the trained novel classifier based on the first feature map.
5 more elements
A system for classifying novel class objects, comprising:
- a processor configured to receive prior knowledge from base classifiers;
- and
- a memory configured to store a program that constructs and trains a novel classifier, by considering the prior knowledge of the base classifiers,
- wherein the processor is further configured to execute the program,
- wherein the program involves learning parameterized weight coefficients of a novel classifier model during the training of the novel classifier and generating parameters of the novel classifier model as a linear combination of parameters of the base classifiers and a preset number of Gaussian random vectors.
2 more elements
Dependencies are inferred from the claim text.