Memorization in 3D Shape Generation: An Empirical Study
Research Paper#3D Generative Models, Memorization, Data Leakage, Shape Generation🔬 Research|Analyzed: Jan 3, 2026 18:34•
Published: Dec 29, 2025 17:39
•1 min read
•ArXivAnalysis
This paper investigates the memorization capabilities of 3D generative models, a crucial aspect for preventing data leakage and improving generation diversity. The study's focus on understanding how data and model design influence memorization is valuable for developing more robust and reliable 3D shape generation techniques. The provided framework and analysis offer practical insights for researchers and practitioners in the field.
Key Takeaways
- •The paper provides a framework to quantify memorization in 3D generative models.
- •Memorization is influenced by data modality, diversity, and conditioning.
- •Model design choices like guidance scale, Vecset length, and augmentation affect memorization.
- •Strategies to reduce memorization without sacrificing generation quality are suggested.
Reference / Citation
View Original"Memorization depends on data modality, and increases with data diversity and finer-grained conditioning; on the modeling side, it peaks at a moderate guidance scale and can be mitigated by longer Vecsets and simple rotation augmentation."