DiverseFlow: Sample-Efficient Diverse Mode Coverage in Flows
This addresses the need for sample-efficient diversity in flow-based generative models for applications like image synthesis, though it is incremental as it builds on existing flow models.
The paper tackles the problem of inefficient diverse sample generation in flow-based models by introducing DiverseFlow, a training-free method that uses a determinantal point process to couple samples and improve diversity under a fixed budget, achieving more variations with fewer samples in tasks like text-guided image generation and inpainting.
Many real-world applications of flow-based generative models desire a diverse set of samples that cover multiple modes of the target distribution. However, the predominant approach for obtaining diverse sets is not sample-efficient, as it involves independently obtaining many samples from the source distribution and mapping them through the flow until the desired mode coverage is achieved. As an alternative to repeated sampling, we introduce DiverseFlow: a training-free approach to improve the diversity of flow models. Our key idea is to employ a determinantal point process to induce a coupling between the samples that drives diversity under a fixed sampling budget. In essence, DiverseFlow allows exploration of more variations in a learned flow model with fewer samples. We demonstrate the efficacy of our method for tasks where sample-efficient diversity is desirable, such as text-guided image generation with polysemous words, inverse problems like large-hole inpainting, and class-conditional image synthesis.