Method for generating picture of pedestrian with arbitrary pose

Abstract

The invention discloses a generative adversarial network-based method for generating a picture of a pedestrian with an arbitrary pose. Only by inputting a character picture, the pose of the pedestrianin the picture and a thermodynamic image of a target transfer pose to a generator network, a picture of the same character with a previous target pose can be obtained. A pose attention mechanism is introduced in a generator of the method, and two different discriminators are adopted for performing appearance consistency and pose consistency discrimination, so that multiple complex conditions of image deformation, geometric conversion, visual angle shift and the like can be met. Moreover, end-to-end training can be realized. According to the method for generating the picture of the pedestrianwith the arbitrary pose, an innovation is performed based on an existing technical idea, and a novel structure is adopted for performing network model building, so that compared with a conventional method, the synthesized picture is realer and more natural, and high practical application values are achieved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and more specifically relates to a method for generating pictures of pedestrians with arbitrary postures. Background technique [0002] Algorithms that generate more realistic-looking, natural-looking images are becoming increasingly popular in the field of computer vision, driven by the growing demand for high-quality synthetic images in real life. And character pose transfer is a very active topic in this field. The generation method of pedestrian pictures with arbitrary poses based on the pose transfer of people has many valuable applications, for example, using pedestrian pictures of different poses as a certain frame of the video, so as to use the sequence of pose pictures of the same person to synthesize the video. As another example, the method of generating a picture of a person with any pose is used as a means of data augmentation, which is especially useful in tasks such as sec...

AI Technical Summary

Problems solved by technology

However, generating high-quality pedestrian pictures with arbitrary poses is a challenging task: since pedestrian pictures in natural scenes are often affected by factors such as pose deformation, geometric transformation, and viewing angle uncertainty, using these previous The network structure often cannot achieve the desired effect

Images