TFTF: Training-Free Targeted Flow for Conditional Sampling

TL;DR

This paper introduces a training-free conditional sampling method for flow matching models that employs importance sampling and resampling techniques, enabling high-quality, diverse sample generation without additional training.

Contribution

The authors develop a novel training-free framework using importance sampling and stochastic flows, with theoretical guarantees and superior performance on image generation tasks.

Findings

Outperforms existing methods on MNIST and CIFAR-10

Provides theoretical guarantees of asymptotic accuracy

Effective in high-dimensional, multimodal settings like CelebA-HQ

Abstract

We propose a training-free conditional sampling method for flow matching models based on importance sampling. Because a na\"ive application of importance sampling suffers from weight degeneracy in high-dimensional settings, we modify and incorporate a resampling technique in sequential Monte Carlo (SMC) during intermediate stages of the generation process. To encourage generated samples to diverge along distinct trajectories, we derive a stochastic flow with adjustable noise strength to replace the deterministic flow at the intermediate stage. Our framework requires no additional training, while providing theoretical guarantees of asymptotic accuracy. Experimentally, our method significantly outperforms existing approaches on conditional sampling tasks for MNIST and CIFAR-10. We further demonstrate the applicability of our approach in higher-dimensional, multimodal settings through text-to-image generation experiments on CelebA-HQ.