Adjustment method and device of face recognition model and storage medium

Abstract

The embodiment of the invention discloses an adjustment method and device of a face recognition model and a storage medium. The embodiment of the invention can obtain an original sample set used for training the face recognition model and an adjustment sample set (a minority population sample set) used for adjusting the model, respectively train the face recognition model according to the same, then determine a similarity distribution curve of the original sample set and a similarity distribution curve of the adjustment sample set, and then adjust the face recognition model through convergenceon the similarity distribution curve of the original sample set and the similarity distribution curve of the adjustment sample set. The scheme fine-tunes the face recognition model on the premise ofnot increasing face recognition system architecture complexity and not impacting majority population face recognition accuracy, and thus facilitates fast improvement of minority population face recognition accuracy.

Description

technical field [0001] The invention relates to the field of communication technology, in particular to a face recognition model adjustment method, device and storage medium. Background technique [0002] In recent years, face recognition technology has been greatly developed. However, for minority groups, such as Uighurs, the elderly, or blacks, most of the existing face recognition systems often have low recognition accuracy, which is specifically manifested in the system calculation. The facial feature similarity among the minority groups of different people is too high, resulting in too many false alarms, which seriously affects the use effect of the face recognition system in some scenarios. [0003] In response to the above problems, several methods have been proposed to improve the face recognition accuracy of minority groups to reduce the false alarm rate of minority groups. For example, retrain the face recognition model by increasing the sample proportion of minor...

AI Technical Summary

Problems solved by technology

However, in the process of research and practice of the prior art, the inventors of the present invention found that if a large number of minority population samples are re-collected to train the face recognition model, it will take a lot of time, and the training samples of the minority population A large increase may affect the recognition accuracy of the face recognition system for most groups of people, and if different alarm thresholds are set for different groups of people, a large adjustment of the system architecture is required, which greatly increases the complexity of the system and also affects the entire face. The recognition accuracy of the recognition system is affected

Images