Recognition and training method and apparatus

A processor implemented training method to train a recognition model comprising a recognition model neural network, comprising:

• grouping training data into a plurality of groups by training, with supervision, a first model into the recognition model neural network;
• excluding group representative information from group data belonging to each of the plurality of groups;
• clustering the group data from which the group representative information is excluded into a plurality of clusters by training, without supervision, a second model into the recognition model neural network;
• and
• grouping cluster data belonging to each of the plurality of clusters into a plurality of sub-groups to train the recognition model by training, with supervision, a third model into the recognition model neural network.

A recognition apparatus, comprising:

• a memory configured to store a recognition model;
• and
• a processor configured to
• calculate feature data corresponding to input data from the input data based on the recognition model,
• identify a cluster to which the input data belongs, and
• recognize the input data based on the feature data and the cluster to which the input data belongs, and
• in response to the cluster to which the input data belongs being different from a cluster to which data to be verified belongs, the processor is further configured to
• convert the input data to the cluster to which the data to be verified belongs by applying a cluster difference between the cluster to which the input data belongs and the cluster to which the data to be identified belongs, and
• recognize the input data based on similarity between feature data of the converted input data and the feature data of the data to be verified exceeding a threshold.

A processor implemented recognition method, comprising:

• calculating feature data corresponding to an input image including an object based on a recognition model from the input image;
• identifying a cluster to which the object included in the input image belongs based on the feature data,
• wherein the identifying comprises obtaining a cluster to which a registered image corresponding to a registered user belongs and obtaining a virtual registered image corresponding to at least one of remaining clusters excluding the cluster to which the registered image belongs;
• and
• recognizing the object based on the feature data, the cluster to which the object included in the input image belongs, and the virtual registered image.