Search papers, labs, and topics across Lattice.
This paper introduces Low-Pass Flow Matching, a novel variant of Flow Matching that utilizes an operator-modulated interpolant to align with the frequency characteristics of natural data. By incorporating a time-varying spectral bias that shifts from the source spectrum to a frequency-decaying bias, the method enhances the efficiency of unconditional image generation tasks. Experimental results demonstrate that when combined with adaptive ODE solvers, Low-Pass Flow Matching significantly improves sample quality and reduces sampling costs compared to traditional approaches.
Low-Pass Flow Matching reduces sampling costs while enhancing image quality by aligning flow matching with the natural frequency decay of data.
Flow Matching typically relies on white noise sources, a choice often misaligned with the power spectra of natural data, which tend to decay with frequency. To address this, we introduce Low-Pass Flow Matching, a variant of Flow Matching based on an operator-modulated interpolant. This formulation induces a time-varying spectral bias that transitions from the source spectrum to a frequency-decaying bias as the path approaches the data. We validate our method on unconditional image generation tasks, including the scientific Galaxy10 dataset. Empirically, we show that our method is particularly effective when paired with adaptive ODE solvers, where it improves or preserves sample quality while substantially reducing sampling cost compared to standard baselines.
