Attitude transformation data processing method, device, computer equipment and storage medium

Abstract

The present application relates to a pose transformation data processing method, device, computer equipment, and storage medium, involving artificial intelligence image processing technology, including: obtaining the source image and the target three-dimensional pose, and reconstructing the three-dimensional image including voxel category information based on semantic segmentation Segment the voxels, project the 3D segmented voxels to obtain the corresponding 2D segmentation map of the target pose, label the objects in the 2D segmentation map of the target pose based on the category information to obtain the component category; obtain the 2D pose of the target corresponding to the 3D pose, extract The source image, target pose 2D segmentation map and target 2D pose features are synthesized into an intermediate-scale transformed image; the source image, 3D segmented voxels, target 2D pose and transformed image are respectively cut to obtain the component layer data of each object part , to perform component synthesis on the component layer data of each object component respectively to generate a component image; fuse the transformed image and the component image to obtain the target pose image, and improve the quality of the pose transformed image.

Description

technical field [0001] The present application relates to the field of image processing, in particular to a pose transformation data processing method, device, computer equipment and storage medium. Background technique [0002] Pose transformation is an emerging research technique in image synthesis that aims to transform a given image of an object into a target pose while maintaining the consistency of the object's appearance. [0003] In related technologies, deep learning technology can be used to solve the pose transformation problem, and the given image features can be deformed to the target pose image through the space transformation module. However, when the visual features change due to the large pose interval, it will be difficult to obtain satisfactory results. , also cannot solve the occlusion problem, resulting in low-quality pose-transformed images. Contents of the invention [0004] Based on this, it is necessary to address the above technical problems and ...

AI Technical Summary

Problems solved by technology

[0003] In related technologies, deep learning technology can be used to solve the pose transformation problem, and the given image features can be deformed to the target pose image through the space transformation module. However, when the visual features change due to the large pose interval, it will be difficult to obtain satisfactory results. , also cannot solve the occlusion problem, resulting in low quality of the pose transformation image

Images