A method and system for generating customized action images based on a general action space

By constructing a general action semantic space and modulating the coefficients of the action basis using a multilayer perceptron, the problem of inconsistent action image generation in existing technologies is solved, enabling customized action image generation that can be reused in different contexts and roles, thus improving generation performance and consistency.

CN119723230BActive Publication Date: 2026-07-14ZHEJIANG UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
ZHEJIANG UNIV
Filing Date
2024-11-01
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively generate customized motion images, resulting in generated images containing information beyond the motion itself. This makes it impossible to reuse motion information across different contexts and roles, and also limits generation performance.

Method used

By constructing a general action semantic space, modulating the coefficients of the action basis using a multilayer perceptron, and introducing action similarity loss, the decoupling of action and actor features is achieved, generating highly customized action images.

Benefits of technology

It improves the performance of motion image generation, ensures consistency between generated images and descriptive text and reference motion images, adapts to different characters and contexts, supports the reuse of motion information in multiple application scenarios, and has good scalability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119723230B_ABST
    Figure CN119723230B_ABST
Patent Text Reader

Abstract

The application discloses a kind of based on general action space's customized action image generation method and system, the method of the application includes three key steps: first, a set of representative action phrase is based on and constructs general action semantic space;Second, imitate customized action in general action semantic space;Finally, the semantic similarity of customized action is optimized by action similarity loss, to generate accurate, context-independent customized action image in different situations, while maintaining the identity consistency of different subjects, including animals, humans and even customized characters.The method of the application is particularly suitable for text-guided diffusion model's few-shot action image synthesis, which constructs a general action semantic space to address the challenge of decoupling action from other semantic features such as the appearance of the person in the reference action image in the case of few-shot images.In addition, the superiority of the method of the application in generating customized action images is verified through experiments.
Need to check novelty before this filing date? Find Prior Art