Target recognition network training method, device, computer equipment and storage medium

Abstract

The present invention relates to a training method, device, computer equipment and storage medium for a target recognition network. The method obtains the weights corresponding to the eigenvalues ​​of each sample image in the sample set, and then according to the distance difference between the unknown weights corresponding to the eigenvalues ​​of each sample image and the known weights corresponding to the eigenvalues ​​of each sample image, and the preset The boundary angle determines the classification margin parameters corresponding to the eigenvalues ​​of each sample image; then further constructs a loss function according to the weights and classification margin parameters, and finally uses the constructed loss function to train the target recognition network. The loss function constructed by the above method can improve the network quality of the target recognition network trained by the loss function, thereby improving the face recognition accuracy when using the trained target recognition network for face recognition.

Description

technical field [0001] The present application relates to the technical field of face recognition, and in particular, to a training method, device, computer equipment and storage medium for a target recognition network. Background technique [0002] With the development of face recognition technology, the application of deep neural network models to various types of face recognition fields has received extensive attention, such as face verification, face retrieval, face clustering and other fields. The quality of the above deep neural network model directly affects the results of the face recognition task. Therefore, how to improve the quality of the deep neural network model through training has become a more concerned issue. [0003] At present, for the training process of the deep neural network model, the commonly used classification learning scheme is to use the loss function of Margin Softmax loss to optimize the classification model. Among them, Softmax loss includes ...

AI Technical Summary

Problems solved by technology

[0004] However, the above-mentioned classification learning method has the problem that the quality of the deep neural network model obtained by training is low, and then the recognition accuracy is low when using the deep neural network model for face recognition.

Images