数据采集与处理2026，Vol.41Issue(2)：543-565,23.DOI:10.16337/j.1004-9037.2026.02.013
三维人脸生成技术综述
A Survey on 3D Face Generation Technology
摘要
Abstract
In recent years,benefiting from the rapid development of computer vision and graphics,3D face generation technology has achieved significant breakthroughs;3D vision technologies,such as digital avatar creation,have become increasingly popular on the internet,attracting extensive attention from both academia and industry.This generation technology synthesizes realistic multi-view face images by reconstructing geometric structures and texture details from explicit or implicit underlying representations.3D face generation technology has sparked many related entertainment and interactive applications,such as using attribute editing technology to modify facial features via text descriptions,or using talking head generation technology to drive a static portrait to generate a talking video.However,early technologies based on linear parametric models suffered from poor realism and detail performance.And the subsequently emerging implicit neural representation technologies,while significantly improving visual quality,face the challenges of high computational costs and difficulty in achieving real-time interaction,which have brought great limitations to practical deployment and application.In order to overcome the contradiction between speed and quality,numerous scholars have conducted in-depth research on novel representations based on explicit Gaussian primitives and generative models based on probabilistic diffusion,and have proposed a series of hybrid generation methods from different perspectives.However,due to problems such as difficulty in generalizing from small sample data,incomplete modeling of full-head physical structures,and insufficient consistency in dynamic driving,there is still a long way to go for generation technology on the path to becoming fully photorealistic and capable of real-time interaction.In fact,research on 3D face generation and driving technology is still in a developmental stage,and the connotations and extensions of its technology are rapidly updating and iterating.This review provides a systematic summary of the main research works to date,along with a brief analysis of the limitations of current technologies.It also explores potential challenges and future directions for 3D face generation and application technologies,offering a guidance for future research.关键词
三维人脸生成/三维可变形模型/神经辐射场/三维高斯泼溅/扩散模型/生成对抗网络Key words
3D face generation/3D morphable models/neural radiance field/3D Gaussian splatting/diffusion models/generative adversarial network分类
信息技术与安全科学引用本文复制引用
王伟,何一康,魏云超,赵耀..三维人脸生成技术综述[J].数据采集与处理,2026,41(2):543-565,23.基金项目
国家自然科学基金(62372033) （62372033）
北京市自然科学基金"海淀联合-重点"项目(L252025). National Natural Science Foundation of China(No.62372033) （L252025）
Natural Science Foundation of Beijing(No.L252025). （No.L252025）
