Shape Translation of Dental Point Clouds
Keywords: Shape Translation, Generative Modeling, Point Clouds
Abstract: Unpaired shape-to-shape translation remains a largely unexplored area, particularly in three-dimensional contexts. This paper explores its potential in dentistry, focusing on the translation of point cloud representations of teeth between young and old patients. We propose a novel approach that combines the latent overcomplete GAN framework with dual diffusion implicit bridges (DDIB) to enhance shape translations improving the applicability of these models in dental contexts. DDIB, a diffusion-based approach leveraging optimal transport properties, demonstrates significant improvements in generating more diverse and cycle-consistent samples that better resemble the target distribution. While these advancements show promise, further research is necessary to develop an autoencoder that balances high reconstruction accuracy with effective shape translation, addressing the unique challenges of dental morphology. Our findings establish a foundation for future research and applications in dentistry, potentially enabling personalized treatments and proactive interventions for various dental conditions.
Submission Number: 32
Loading