Stochastic Perturbations Improve Distribution-to-Distribution Generative Models

Shiye Su, Yuhui Zhang, Linqi Zhou, Rajesh Ranganath, Serena Yeung-Levy; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Findings, 2026, pp. 3965-3974

Modeling transformations between data distributions is a fundamental scientific problem, with applications in fields such as drug discovery and evolutionary simulation. While flow matching offers a natural framework for this task, prior work has primarily focused on generating data from noise, while the more general distribution-to-distribution setting is underexplored. We find that in the latter case, where the source is also a data distribution to be learned from finite samples, standard flow matching break down due to sparse supervision. To mitigate this, we introduce a simple and computationally efficient method that injects stochasticity during training by perturbing source samples and flow interpolants. On five diverse imaging tasks spanning biology, radiology, and astronomy, our method significantly improves generation quality, surpassing baselines by 9 FID points on average. Our approach also reduces the transport cost between input and generated samples to better highlight the true effect of the transformation, establishing flow matching as a more practical tool for simulating diverse distribution transformations. Our code is available at https://github.com/ssu53/StochasticPerturbations.