A Survey on 3D Face Generation Technology
CSTR:
Author:
Affiliation:

1Institute of Information Science, Beijing Jiaotong University, Beijing 100044, China;2Visual Intelligence + X International Cooperation Joint Laboratory of the Ministry of Education, Beijing 100044, China

Clc Number:

TP391.41

Fund Project:

National Natural Science Foundation of China (No.62372033); Natural Science Foundation of Beijing (No.L252025).

  • Article
  • |
  • Figures
  • |
  • Metrics
  • |
  • Reference
  • |
  • Related
  • |
  • Cited by
  • |
  • Materials
  • |
  • Comments
    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.

    Reference
    Related
    Cited by
Get Citation

WANG Wei, HE Yikang, WEI Yunchao, ZHAO Yao. A Survey on 3D Face Generation Technology[J]. Journal of Data Acquisition and Processing,2026,(2):543-565.

Copy
Related Videos

Share
Article Metrics
  • Abstract:
  • PDF:
  • HTML:
  • Cited by:
History
  • Received:January 09,2026
  • Revised:February 26,2026
  • Adopted:
  • Online: April 15,2026
  • Published:
Article QR Code