Apparatus and method for training classification model and apparatus for performing classification by using classification model

A method for training a classification model, comprising:

• a feature extraction step of, setting, with respect to each training set of a first predetermined number of training sets, a corresponding number of feature extraction layers respectively, and extracting, with respect to each sample image, features of the sample image respectively by using feature extraction layers set with respect to a training set to which the sample image belongs, to obtain extracted features of the sample image, wherein at least two of the first predetermined number of training sets at least partially overlap;
• a feature fusion step of, setting, with respect to each training set of the first predetermined number of training sets, a corresponding number of feature fusion layers respectively, and performing, with respect to each sample image, a fusion on the extracted features of the sample image by using feature fusion layers set with respect to a training set to which the sample image belongs, to obtain fused features of the sample image;
• and
• a loss determination step of, setting, with respect to each training set of the first predetermined number of training sets, a loss determination layer respectively, calculating, with respect to each sample image, a loss function of the sample image based on the fused features of the sample image by using a loss determination layer set with respect to a training set to which the sample image belongs, and training the classification model based on the loss function,
• wherein the first predetermined number of training sets share at least one layer of the feature fusion layers and the feature extraction layers which are set with respect to each training set respectively.

An apparatus for performing classification by using a classification model, comprising:

• a second feature extraction unit configured to extract, with a corresponding number of feature extraction layers in the classification model which are set with respect to each training set of a first predetermined number of training sets respectively, features of an image to be classified respectively, to obtain extracted features of the image to be classified;
• a second feature fusion unit configured to perform, by using a corresponding number of feature fusion layers in the classification model which are set with respect to each training set of the first predetermined number of training sets respectively, a fusion on the extracted features of the image to be classified, to obtain fused features of the image to be classified;
• and
• a second loss determination unit configured to calculate, with loss determination layers in the classification model which are set with respect to the first predetermined number of training sets respectively, a loss function of the image to be classified based on the fused features of the image to be classified, and to classify the image to be classified based on the loss function,
• wherein in the classification model, the first predetermined number of training sets share at least one layer of the feature fusion layers and the feature extraction layers which are set with respect to each training set respectively.