Model training, face recognition model updating method, electronic device and storage medium

Abstract

The embodiment of the present invention relates to the field of face recognition, and discloses a model training, a face recognition model updating method, an electronic device, and a storage medium, wherein the model training includes: using any layer of the feature extraction network in the new face recognition model to output The feature map of the feature map is input, and a feature map branch network is added to form an intermediate model; the old model of face recognition is used as a teacher model, and the intermediate model is a student model, and the feature map branch network is subjected to distillation training, and the training is obtained. The intermediate model. This solution can map the feature space of the new face recognition algorithm to the feature space of the old face recognition algorithm through the mapping matrix, and gradually update the face feature database while ensuring the security of face recognition while avoiding the problem of face privacy. In this way, the user-side non-sensing upgrade is realized and the user experience is improved.

Description

technical field [0001] The invention relates to the field of face recognition, in particular to a model training, a method for updating a face recognition model, electronic equipment and a storage medium. Background technique [0002] As one of the hot issues in the field of computer vision, face recognition has entered a deep water period with the development of deep learning technology. During the use of face recognition products, developers will continue to upgrade face recognition algorithm technology to solve the pain points of users during use. At present, the face recognition algorithm in the industry updates the face feature database mainly adopts the following mainstream strategies: [0003] 1. The background / device locally retains the original registration picture when the user registers for the first time. When the face recognition algorithm is updated, use the new face recognition algorithm to extract features from all the saved registration pictures, so as to r...

AI Technical Summary

Problems solved by technology

However, this method is cumbersome, the user experience is poor, and it cannot highlight the advantages of face recognition products

In addition, the prerequisite for this method to be implemented is that from the algorithm side, it is necessary to ensure that the feature space of the new and old algorithms is as close as possible, otherwise it may cause serious face misrecognition and bring unpredictable risks to the security of face recognition.

Due to the existence of this prerequisite, additional restrictions have been imposed on subsequent algorithm updates, resulting in no significant improvement in the user experience of the updated algorithm

Images