Search papers, labs, and topics across Lattice.
The paper introduces MixFlow, a training strategy for rectified flows that reduces generative path curvature by training on a mixture of a fixed unconditional distribution and a conditioned source distribution (κ-FC). κ-FC is a novel formulation that conditions the source distribution on a signal κ to better align it with the data distribution, moving beyond the standard Gaussian assumption. Experiments demonstrate that MixFlow improves generation quality by 12% in FID compared to standard rectified flow and 7% compared to previous baselines under a fixed sampling budget, while also accelerating training convergence.
Rectified flows get a speed boost and quality bump by ditching the standard Gaussian source distribution and mixing in a conditioned distribution, leading to less curved generative paths.
Diffusion models and their variations, such as rectified flows, generate diverse and high-quality images, but they are still hindered by slow iterative sampling caused by the highly curved generative paths they learn. An important cause of high curvature, as shown by previous work, is independence between the source distribution (standard Gaussian) and the data distribution. In this work, we tackle this limitation by two complementary contributions. First, we attempt to break away from the standard Gaussian assumption by introducing κ-FC, a general formulation that conditions the source distribution on an arbitrary signal κ that aligns it better with the data distribution. Then, we present MixFlow, a simple but effective training strategy that reduces the generative path curvatures and considerably improves sampling efficiency. MixFlow trains a flow model on linear mixtures of a fixed unconditional distribution and a κ-FC-based distribution. This simple mixture improves the alignment between the source and data, provides better generation quality with less required sampling steps, and accelerates the training convergence considerably. On average, our training procedure improves the generation quality by 12\% in FID compared to standard rectified flow and 7\% compared to previous baselines under a fixed sampling budget. Code available at: https://github.com/NazirNayal8/MixFlow{https://github.com/NazirNayal8/MixFlow}
