Classification model training method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a training method of a classification model. The method comprises the steps of obtaining different types of training data to train the classification model; in the current iterative training process, calculating a first center distance from the current sample to the category box of the corresponding category; obtaining a first loss function of the classification model in the previous iterative training process, wherein the first loss function comprises a first category box parameter determined in the previous iterative training process; updating a first category box parameter in the first loss function according to the first center distance and the first category box parameter to obtain a second category box parameter, and determining a second loss function of a category corresponding to the current sample based on the second category box parameter; determining a dynamic loss function corresponding to each category through the second loss function of each category; and training the classification model according to the dynamic loss function. By means of the method, the training efficiency of the classification model and the classification recognition accuracy of the classification model are improved.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a training method, device, electronic equipment and storage medium for a classification model. Background technique [0002] In the training process of the classification model, it is necessary to use sample data as input, so that the classification model can learn the classification and recognition of the sample data under supervision. In order to make the classification model have higher classification accuracy, the distance between each category can be increased, and the distance between each sample in the same category can be reduced. The usual practice is to add a hyperparameter margin to the same category in the loss function. The samples are framed, and the in-class samples that exceed the framed range are punished, so that in the subsequent training, the in-class samples are closer to the framed range. However, in the actual training task, it is difficult to set a ...

AI Technical Summary

Problems solved by technology

However, in the actual training task, it is difficult to set a margin value suitable for this batch of data, or a large number of parameter adjustment experiments and trade-offs between the accuracy of each category are required to debug a set of data for this data set. Margin value, which not only wastes a lot of energy and time to "trial and error", but eventually leads to a very small probability of finding a suitable margin value

Therefore, the existing method of obtaining the hyperparameter margin is complex and difficult, which makes the training efficiency of the classification model low

Images