Special effect model training method and device and special effect generating method and device

Abstract

The invention relates to a special effect model training method and device and a special effect generation method and device. The special effect model training method comprises the steps that a training sample set is acquired, wherein the training sample set comprises a plurality of training images; the training images and a first mask graph are input into a special effect model to obtain an output image and a second mask graph after a special effect is added to a preset object in the training images, wherein the first mask graph is the mask graph of the preset object in the training images, and the second mask graph is the mask graph of the preset object after the special effect is added to the output image; the value of a target loss function is determined based on the training images, the output image, the first mask graph and the second mask graph; and the parameters of the special effect model are adjusted based on the value of the target loss function, and the special effect model is trained. According to the invention, the problem that the GAN special effect is not ideal in color, texture and other attributes in the prior art can be solved.

Description

technical field

[0001] The present disclosure relates to the field of image processing, and in particular to a method and device for training a special effect model, and a method and device for generating a special effect. Background technique

[0002] At present, in common short video applications or camera applications, Generative Adversarial Networks (GAN for short) special effects have become more and more common, such as special effects such as getting old, changing children, comic faces, hand-painting, etc. Or the change of style (such as face attribute or style) is very natural and real, so it is very popular among users.

[0003] In related technologies, generally based on image domain (domain) migration schemes, such as cyclegan, stargan and other schemes, the properties of the object to be processed are manipulated to achieve changes in attributes. However, the effect of this scheme on attributes such as color and texture not ideal. Take the hair growth effect as...

Examples

Embodiment Construction

[0046] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0047] It should be noted that the terms "first" and "second" in the specification and claims of the present disclosure and the above drawings are used to distinguish similar objects, but not necessarily used to describe a specific sequence or sequence. It is to be understood that the data so used are interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein can be practiced in sequences other than those illustrated or described herein. The implementations described in the following examples do not represent all implementations consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with aspect...

technical field

[0001] The present disclosure relates to the field of image processing, and in particular to a method and device for training a special effect model, and a method and device for generating a special effect. Background technique

[0002] At present, in common short video applications or camera applications, Generative Adversarial Networks (GAN for short) special effects have become more and more common, such as special effects such as getting old, changing children, comic faces, hand-painting, etc. Or the change of style (such as face attribute or style) is very natural and real, so it is very popular among users.

[0003] In related technologies, generally based on image domain (domain) migration schemes, such as cyclegan, stargan and other schemes, the properties of the object to be processed are manipulated to achieve changes in attributes. However, the effect of this scheme on attributes such as color and texture not ideal. Take the hair growth effect as...

Examples

Embodiment Construction

[0046] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0047] It should be noted that the terms "first" and "second" in the specification and claims of the present disclosure and the above drawings are used to distinguish similar objects, but not necessarily used to describe a specific sequence or sequence. It is to be understood that the data so used are interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein can be practiced in sequences other than those illustrated or described herein. The implementations described in the following examples do not represent all implementations consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with aspect...

Images