Domain self-adaptive method and device based on comparative adversarial learning

Abstract

A domain self-adaptive method and device based on comparative adversarial learning are provided. On source domain data, a loss function Lcls(xs, ys) is used for training a whole network model, parameters in a feature extractor are fixed, only classifiers C1 and C2 are updated, classification loss of the classifiers is minimized, target domain sample discrimination differences of the classifiers are maximized, the parameters in the classifiers C1 and C2 are fixed, Ldis is used for updating parameters in the feature extractor, and a self-adaptive loss item is reserved in the step. The classifier C1 and the classifier C2 respectively use features of different data enhancement modes, so that diversity of the classifiers is guaranteed, the double classifiers can more efficiently find samples at classification boundaries, features learned by the model contain more effective information, so that the problem of self-adaption in the unsupervised field is well solved. On the basis of a traditional double-classifier-based confrontation method, the disclosed method takes decision boundaries of classifiers on a target domain into account and also pays close attention to inter-domain differences.

Description

technical field [0001] The present invention relates to the technical field of computer vision, in particular to a field adaptive method based on contrastive confrontation learning and a field adaptive device based on contrastive confrontation learning. Background technique [0002] The invention focuses on solving the image classification problem of unsupervised domain self-adaptation, and combines the deep network and domain self-adaptive problems. By processing the characteristics of the samples in the feature space, the difference between the distribution of the source domain and the target domain is reduced, so that the knowledge learned in the source domain can also be applied to the target domain. Deep unsupervised domain adaptation is a research field that has attracted much attention, and a large number of scholars have participated in the research work in this field. At present, deep unsupervised domain adaptation methods can be mainly divided into three categorie...

AI Technical Summary

Problems solved by technology

Such methods require encoders and decoders

Images