Virtual human body image generation method, system and device and medium

Abstract

The invention discloses a virtual human body image generation method, system and device and a medium, and the method comprises the steps: inputting a source human body image and a target posture image into a pre-trained virtual human body image generation network, and obtaining a target posture human body image, wherein the virtual human body image generation network is a convolutional neural network, and comprises: an encoder, which is used for inputting a source human body image and a target posture image, and encoding to obtain source human body features and target human body features, a structure-based appearance generation module, which is used for inputting source human body features and target human body features and updating the source human body features and the target human body features to obtain the updated source human body features and target human body features; and a decoder, which is used for inputting the target human body characteristics output by the structure-based appearance generation module and decoding to obtain a target posture human body image. The virtual human body image generation network under the posture guidance based on the human body structure is utilized, and the vivid human body image with the correct target posture can be generated.

Description

technical field [0001] The invention belongs to the interdisciplinary technical fields of computer vision and computer graphics, and in particular relates to a method, system, device and medium for generating a virtual human body image. Background technique [0002] The goal of the pose-guided virtual human body image task is to generate a new human body image of a target pose for a source human body image and a given target pose image; wherein, the human body pose and target pose of the generated new human body image Consistent, and the body appearance is similar to that of the source body image. This task has many application scenarios, such as film production, virtual reality, data augmentation in action recognition tasks, etc. [0003] At present, the virtual image generation method under the attitude guidance at this stage mainly has the following problems and defects: [0004] Generating a virtual human image of the target pose should take into account both the pose ...

AI Technical Summary

Problems solved by technology

[0004] Generating a virtual human image of the target pose should take into account both the pose consistency and appearance consistency of the generated image; the target pose and the pose in the source human image are usually very different, and it is very complicated. Existing methods usually consider how to combine the source The human body image is deformed to obtain the human body image of the target pose. Since the effectiveness of the deformation cannot be guaranteed, the human body pose of the generated image is easily unclear or even inconsistent with the target pose, so that the quality of the generated human body image of the specified pose is very low.

Images