AI Image Generation Using Pose Transition Matrix
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing human pose transfer technologies based on pose heatmaps struggle to accurately characterize human poses, leading to significant differences between the transferred pose and the target pose, resulting in undesirable image generation effects.
Innovation Solution
An artificial intelligence-based image generation method that uses a model pose to characterize human poses and determines a pose transition matrix based on the model poses of the target object and the reference object in the source and target images, respectively, to accurately transfer the human pose.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If pose heatmap is used to characterize human pose, then the method is simple to implement, but the pose transfer accuracy deteriorates significantly
Solution Approach 1:
The patent changes the parameter representation from pose heatmap (2D spatial distribution) to model pose (3D joint coordinates). This parameter transformation enables more accurate pose characterization by using explicit joint position data instead of heatmap-based probability distributions, directly resolving the accuracy issue while maintaining computational feasibility through standard coordinate transformations.
Solution Approach 2:
The patent introduces a pose transition matrix as an intermediary to bridge the source pose and target pose. This matrix serves as a transformation mediator that maps the difference between source and target model poses, enabling accurate pose transfer by composing the source appearance features with the transition matrix to generate the target pose representation.
2Speed
If pose heatmap-based transfer is used, then the processing speed is fast, but the generated image quality deteriorates
Solution Approach 1:
The patent transforms the parameter representation from pose heatmap to model pose with explicit joint coordinates. This change improves image generation quality by providing more precise pose information that better captures human body structure, while the processing speed is maintained through efficient matrix operations and feature transformations.
Solution Approach 2:
The patent segments the pose representation into distinct joint points and body parts through model pose decomposition. This segmentation allows for more precise control over different body regions during pose transfer, improving overall image generation quality by enabling targeted adjustments to specific pose elements rather than treating the entire pose as a unified heatmap distribution.
Data Source
AI summary
An embodiment of this application discloses an artificial intelligence-based image generation method performed by a computer device. The method includes: acquiring a source image including a target object whose pose is to be transformed, and a target image including a reference object presenting a target pose; determining a pose transition matrix according to a model pose corresponding to the pose of the target object and a model pose corresponding to the target pose of the reference object; extracting a basic appearance feature of the target object from the source image; processing the basic appearance feature based on the pose transition matrix, to obtain a target appearance feature of the target object in the target pose; and generating a target synthetic image of the target object in the target pose based on the target appearance feature.