Training and feature extraction method of feature extraction network based on multiple data sets

Abstract

The invention relates to the technical field of image processing, in particular to a training and feature extraction method of feature extraction network based on multiple data sets, and the training method comprises the steps: obtaining a plurality of sample data sets; sequentially inputting the sample images in any two sample data sets into a feature extraction network to obtain features corresponding to the sample images; performing gradient inversion on the features corresponding to all the sample images to obtain gradient inversion processing results corresponding to the sample images; based on a gradient inversion processing result, determining a bulldozer distance between any two sample data sets; and training the feature extraction network according to the bulldozer distance to obtain a target feature extraction network. On the basis of a gradient inversion processing result, the bulldozer distance between any two sample data sets is calculated, and the bulldozer distance can still reflect the distance between two distributions when the two distributions are not overlapped, so that the gradient dispersion phenomenon occurring in the GRL can be relieved to a great extent.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for training and feature extraction of a feature extraction network based on multiple data sets. Background technique [0002] The rise of deep learning technology has greatly promoted the development of large-scale image retrieval. Deep learning has a great demand for data. The more abundant the data, the better the effect. However, the labeling of data requires a lot of manpower and time, so multiple Dataset joint training has become a more convenient way to increase the size of the data. [0003] However, due to the differences in the background and application scenarios of different data sets, the joint training of multiple data sets will bring about the problem of domain shift and bring noise to feature learning. Some existing methods to solve the problem of domain migration are called domain migration. For example, domain migration is realized based on gr...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the embodiment of the present invention provides a method for training and feature extraction of a feature extraction network based on multiple data sets to solve the problem of gradient dispersion in GRL training

Images