3DGrowthNet: A Deep Learning Model for Synthetic Aging and Conditional Shape Generation Using 3D Facial Meshes - 25.27
Title:
3DGrowthNet: A Deep Learning Model for Synthetic Aging and Conditional Shape Generation Using 3D Facial Meshes
Authors:
Nina CLAESSENS 1,2, Susan WALSH 3, Mark SHRIVER 4, Seth M WEINBERG 5, Paolo M CATTANEO 6, Anthony J PENINGTON 7,8,9, Peter CLAES 1,2,10
1 UZ Leuven, Medical Imaging Research Center, Leuven, Belgium;
2 Dept. of Electrical Engineering, ESAT/PSI, KU Leuven, Belgium;
3 Department of Biology, Indiana University Purdue University Indianapolis, Indianapolis, IN, USA;
4 Department of Anthropology, Pennsylvania State University, State College, PA, USA;
5 Center for Craniofacial and Dental Genetics, University of Pittsburgh, Pittsburgh, PA, USA;
6 Melbourne Dental School, Faculty of Medicine, Dentistry and Health Sciences, The University of Melbourne, Australia;
7 Facial Sciences Research Group, Murdoch Children's Research Institute, Parkville, Australia;
8 Department of Plastic and Maxillofacial Surgery, Royal Children's Hospital, Melbourne, Australia;
9 Department of Pediatrics, University of Melbourne, Melbourne, Australia;
10 KU Leuven, Human Genetics Department, Leuven, Belgium
Keywords:
facial growth, geometric deep learning, synthetic aging, conditional shape generation, age prediction, sex classification, 3D surface scans, mesh synthesis
Abstract:
Accurately modeling facial growth is essential for applications in forensic science, clinical genetics, and developmental biology. We present 3DGrowthNet, a multi-task geometric deep learning framework that performs continuous synthetic aging, age and sex estimation, and conditional shape generation from 3D facial meshes. We introduce a multi-task training strategy that unifies existing synthetic aging frameworks with conditional shape generation and a continuous label embedding mechanism into a single CVAE-GAN architecture. This integration enables the network to disentangle age from identity while learning to generate anatomically plausible faces across the full age range of 0-88 years.
Trained on over 5,000 scans and validated using a smaller longitudinal dataset of 60 children, the model achieves a mean prediction error of ~2 mm, improving age-invariant identification performance by nearly 20%. It also generates realistic, demographically consistent synthetic faces with high coverage (98.7%) and low Minimum Matching Distance, supporting robust data augmentation. Biomedical relevance is demonstrated through simulations of sexual dimorphism across age, revealing expected developmental trends even in sparsely sampled age ranges. Experiments show that the model is capable of generating a wide variety of realistic and demographically consistent 3D faces and supports robust data augmentation across the age spectrum. In addition to its generative capabilities, 3DGrowthNet performs age and sex estimation resulting in a median absolute age estimation error of 2.0 years and an overall sex classification accuracy of 87.7%, with performance varying across age groups. These results confirm that the model effectively encodes biologically relevant demographic information.
3DGrowthNet sets a new benchmark for realistic, demographically informed mesh synthesis and provides a foundation for advancing personalized growth modeling, forensic identification, and clinical assessment of facial dysmorphism
Full paper:
Note:
This paper is currently under review for publication in the 3DBDOY.TECH Journal - Vol. 2, 2025 (jrnl.3dbody.tech).
It will be updated also in this page once the review process is concluded.
Presentation:
VIDEO will be available here in Q3.2026.
VIDEO availble in proceedings (purchase order)
How to Cite (MLA):
N. Claessens et al., "3DGrowthNet: A Deep Learning Model for Synthetic Aging and Conditional Shape Generation Using 3D Facial Meshes", Proceedings of 3DBODY.TECH 2025 - 16th International Conference and Expo on 3D/4D Body Scanning, Data and Processing Technologies, Lugano, Switzerland, 21-22 Oct. 2025, #27, https://doi.org/10.15221/25.27
Details:
Proceedings: 3DBODY.TECH 2025, 21-22 Oct. 2025, Lugano, Switzerland
Paper/Presentation: #27
DOI: https://doi.org/10.15221/25.27
License/Copyright notice
Proceedings: © Hometrica Consulting - Dr. Nicola D'Apuzzo, Switzerland, hometrica.ch.
Authors retain all rights to individual papers, which are licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
The papers appearing in the proceedings reflect the author's opinions. Their inclusion in the proceedings does not necessary constitute endorsement by the editor or by the publisher.
Note: click the + on the top left of the page to open/close the menu.