Action migration model training method and action migration method

Abstract

The invention provides an action migration model training method and an action migration method. The training method comprises the steps: inputting a source image and a driving image into a to-be-trained action migration model, so that the model obtains a source key point coordinate set of the source image and a driving key point coordinate set of the driving image according to initial model parameters; according to the source key point coordinate set and the driving key point coordinate set, acquiring an optical flow graph and a redrawing which are converted from the source key point coordinate set to the driving key point coordinate set, acquiring a hidden layer feature graph of the source image, and obtaining a target image according to the hidden layer feature graph, the optical flow graph and the redrawing, wherein the target image is an image obtained after a to-be-migrated action associated with the driving key point coordinates in the driving image is migrated to the source key point coordinates; calculating a loss function according to the target image, the source image and the driving image, optimizing initial model parameters according to the loss function until a model convergence condition is reached, completing training, and obtaining a target action migration model.

Description

technical field [0001] The present application relates to the field of image processing, and in particular to a method for training an action transfer model and an action transfer method. Background technique [0002] Action transfer is an important image editing technology in the field of image processing. It refers to the transfer of the action of the characters in the driving video to the characters in the source image. On the basis of retaining the appearance of the characters in the source image, the characters in the source image Synchronize motion with the people in the target video. [0003] In existing schemes, a three-dimensional (3-dimensional, 3D) method can be used for motion transfer. First, it is necessary to analyze the pixel data of the source image to obtain the pixel position information of the character image in the source image. According to the pixel position information of the character image, A 3D model of the character is constructed, and then based...

AI Technical Summary

Problems solved by technology

[0004] However, building a character model using 3D modeling requires the use of a large amount of data and involves complex calculations. Correspondingly, in the actual training process of the model, it not only consumes a lot of computing resources, but also affects the training efficiency of the model.

Images