Image processing method and device, equipment and medium

Abstract

The invention discloses an image processing method and device, equipment and a medium, and the method comprises the steps: obtaining sample data, calling an image processing model to convert an annotated real face image Ai into a reconstructed cartoon face image B'i, and converting a non-annotated real face image Ci into a reconstructed cartoon face image B''i; inversely converting the reconstructed cartoon face image B''i into a reconstructed real face image C''i; inversely converting the marked cartoon face image Bi into a reconstructed real face image C'i; obtaining a first loss according to the difference between B'i and Bi; obtaining a first loss according to B'i, B''i and Bi, obtaining a second loss according to B'i, B''i and Bi, obtaining a third loss according to Ci and C'i, obtaining a fourth loss according to the difference between Ci and C''i; and training an image processing model based on the first loss, the second loss, the third loss and the fourth loss. By means of the method, the model training effect can be improved, and the universality and stability of the trained image processing model are improved.

Description

technical field [0001] The present application relates to the field of computer technology, in particular to the field of artificial intelligence technology, and in particular to an image processing method, an image processing apparatus, an image processing device, and a computer-readable storage medium. Background technique [0002] The development of two-dimensional culture among young groups is very rapid. As more and more Generation Z (ie the Internet generation) enter the society, two-dimensional culture will gradually become the mainstream. Therefore, providing two-dimensional customized services for Internet users is in the pan-entertainment industry. important in the context. The two-dimensional customization service is a service that provides Internet users with a service that converts real face images into cartoon face images. [0003] Based on the requirements of the two-dimensional customization service, the conventional image processing model can be trained, an...

AI Technical Summary

Problems solved by technology

Practice has found that the current training of image processing models for realizing two-dimensional customized services includes two mainstream solutions: unsupervised training and supervised training. The unsupervised training solution refers to obtaining a large amount of unlabeled training data from the Internet as sample data. The training scheme for the model, the model trained by this scheme has poor versatility and cannot be customized in cartoon style; the supervised training scheme refers to the scheme for model training using a large amount of labeled data as sample data, but the current cartoon The labeled data in the domain space needs to be drawn by the painter, and the cost of drawing is high, which makes the cost of the supervised training scheme high and the versatility is poor

It can be seen that the current model training effect for the two-dimensional customized service is poor, and the trained model cannot better realize the two-dimensional customized service

Images