Mitigate Replication and Copying in Diffusion Models with Generalized Caption and Dual Fusion Enhancement
This addresses privacy concerns for users of diffusion models by mitigating data replication, though it appears incremental as it builds on known issues of caption insufficiency and training duplication.
The paper tackles the problem of diffusion models replicating training data by introducing a generality score to measure caption generality and using LLMs to generalize captions, then proposing a dual fusion enhancement approach that reduces replication by 43.5% while maintaining generation diversity and quality.
While diffusion models demonstrate a remarkable capability for generating high-quality images, their tendency to `replicate' training data raises privacy concerns. Although recent research suggests that this replication may stem from the insufficient generalization of training data captions and duplication of training images, effective mitigation strategies remain elusive. To address this gap, our paper first introduces a generality score that measures the caption generality and employ large language model (LLM) to generalize training captions. Subsequently, we leverage generalized captions and propose a novel dual fusion enhancement approach to mitigate the replication of diffusion models. Our empirical results demonstrate that our proposed methods can significantly reduce replication by 43.5% compared to the original diffusion model while maintaining the diversity and quality of generations. Code is available at https://github.com/HowardLi0816/dual-fusion-diffusion.