Figure video generation method based on generative adversarial network

Abstract

The invention discloses a generative adversarial network-based figure video generation method. The method comprises the following steps of 1, collecting an original image and a target image; 2, for the collected original images and target images, extracting multi-scale features between a target attitude and the plurality of original images by using a multi-scale feature extraction module; 3, taking the multi-scale features as input of a global-local module, establishing a global corresponding relation between the target attitude features and original image features by using a global module, and then correcting an output result of the global module by using a local module; 4, selecting an original image, reconstructing the original image to a specific size by adopting pooling operation, and performing deformation operation by using the corrected flow field to obtain a final feature map; and 5, mapping the feature map from the feature space to the image space by adopting a texture renderer to obtain a final generated image. The video can be generated according to the target posture while the clothing texture of the original image is kept unchanged.

Description

technical field [0001] The invention belongs to the technical field of computer vision and image processing, and relates to a character video generation method based on a generative confrontation network. Background technique [0002] The generative model is the core of the field of computer vision. In recent years, methods such as GAN and VAE have achieved impressive results in various image-based generative tasks. In contrast, video-based generative tasks have made little progress, especially Generate a video containing character images (also known as animation generation), because in addition to ensuring that each frame of image generated is real, it is also necessary to ensure the timing consistency of the generated video frames. In practical applications, the generation of static images often cannot meet the needs. On the contrary, videos with dynamic properties can provide better user interaction experience. Character video generation not only needs to ensure that the...

AI Technical Summary

Problems solved by technology

In practical applications, the generation of static images often cannot meet the needs. On the contrary, videos with dynamic properties can provide better user interaction experience.

Images