Method, device and equipment for optimizing model and storage medium

Abstract

The invention discloses a method and device for optimizing a model, equipment and a storage medium, and relates to the technical field of augmented reality, deep learning and animation. According to the specific implementation scheme, the method comprises the steps of obtaining initial coordinates of skin vertexes in a skin skeleton model of a virtual image; generating a plurality of parameter vectors for the target skeleton solving model; on the basis of the parameter vectors, executing the following iterative operations for multiple times: on the basis of the parameter vectors, the target skeleton solving model and the initial coordinates of the skin vertexes, determining skeleton driving coefficients for controlling a plurality of skeleton nodes of the skin vertexes; based on each skeleton driving coefficient and a preset loss function, determining a loss function value corresponding to each skeleton driving coefficient; and optimizing a parameter vector of the target skeleton solving model based on each loss function value. According to the implementation mode, a lightweight model can be provided, and the calculation efficiency is improved.

Description

technical field [0001] The present application relates to the field of computer technology, specifically to the field of augmented reality, deep learning, and animation technology, and in particular to methods, devices, equipment and storage media for optimizing models. Background technique [0002] With the continuous development of technology, augmented reality technology emerges at the historic moment, which can simulate the expression or posture of the real image by driving the virtual image, increasing the user's fun of use. A typical technology is Photo-To-Avatar, which generates personalized 3D animated images based on user pictures. The existing Photo-To-Avatar technology suffers from a large amount of data redundancy, which leads to problems such as excessive network traffic in practical applications and difficulty in deploying on mobile terminals. For the purpose of lightweight, it is necessary to use the skinned bone model to achieve model deformation. Contents...

AI Technical Summary

Problems solved by technology

The existing Photo-To-Avatar technology suffers from a large amount of data redundancy, which leads to problems such as excessive network traffic in practical applications and difficulty in deploying on mobile terminals.

Images