Generative stream model-based human motion style migration method and system

Abstract

The invention provides a human motion style migration method and system based on a generative flow model, and the method comprises the steps: obtaining a preset motion style sequence and a motion content sequence, carrying out the control signal extraction and normalization processing of the motion content sequence, and obtaining motion content input data; inputting the preset exercise style sequence into a trained human body exercise style migration model to obtain a hidden code corresponding to the preset exercise style sequence; inputting the hidden codes and the motion content input data into a trained human body motion style migration model, obtaining a motion content sequence after style migration, and obtaining the trained human body style migration model by training a generation stream model through a sample motion content sequence. According to the method, the stream model is generated to extract the hidden codes, so that the exercise style characteristics are efficiently and accurately extracted, the exercise content is kept unchanged while the exercise style is migrated, and more accurate human body exercise can be synthesized.

Description

technical field [0001] The invention relates to the technical fields of computer vision and artificial intelligence, in particular to a method and system for transferring human movement styles based on a generated flow model. Background technique [0002] The existing human motion style transfer method requires paired motion data, where paired means that the motion data belongs to the same motion content, but the motion style is different, or does not depend on the paired motion data, but requires deep learning methods Perform supervised training to learn a motion transfer model. [0003] Existing methods are based on paired motion data and often require complex data preprocessing, for example, the registration of human motion data in stages and each stage separately; or movement style transfer is limited to paired motion data; or supervised training is required In the deep learning model, each training data needs to be labeled during the training process, which increases t...

AI Technical Summary

Problems solved by technology

[0005] In view of the difficulty and low accuracy of human body movement style transfer training in the prior art, the present invention provides a human body movement style transfer method and system based on a generative flow model

Images