改进的 MeanFlow：论快进生成模型的挑战
Improved Mean Flows: On the Challenges of Fastforward Generative Models
把 MeanFlow 的训练目标改写为对瞬时速度 v 的回归（网络仍预测平均速度 u），并把引导尺度作为显式条件变量做 in-context 条件化，使一步生成模型在测试时可自由调节引导强度；做少步/一步扩散或流匹配采样的人可直接复用这一目标重参数化与条件化引导设计。
原文依据 we recast the objective as a loss on the instantaneous velocity v, re-parameterized by a network that predicts the average velocity u
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论MeanFlow（MF）近来已被确立为一种一步生成建模框架。然而，其“快进”特性在训练目标和引导机制两方面都带来了关键挑战。首先，原始 MF 的训练目标不仅依赖于底层真值场，还依赖于网络自身。为解决该问题，我们将目标重新表述为对瞬时速度 v 的损失，并由一个预测平均速度 u 的网络进行重参数化。我们的重新表述得到了一个更标准的回归问题，并提升了训练稳定性。其次，原始 MF 在训练时固定无分类器引导尺度，牺牲了灵活性。我们通过将引导表述为显式条件变量来解决该问题，从而在测试时保留灵活性。多样的条件通过上下文条件化进行处理，这减小了模型规模并有利于性能。总体而言，我们的改进版 MeanFlow（iMF）方法完全从零开始训练，在 ImageNet 256x256 上以单次函数评估（1-NFE）取得 1.72 的弗雷歇初始距离。iMF 大幅优于此类先前的方方法，并在不使用蒸馏的情况下缩小了与多步方法的差距。我们希望本工作能进一步推动快进生成建模成为一种独立范式。
英文原文摘要 1335 字符
MeanFlow (MF) has recently been established as a framework for one-step generative modeling. However, its "fastforward" nature introduces key challenges in both the training objective and the guidance mechanism. First, the original MF's training target depends not only on the underlying ground-truth fields but also on the network itself. To address this issue, we recast the objective as a loss on the instantaneous velocity v, re-parameterized by a network that predicts the average velocity u. Our reformulation yields a more standard regression problem and improves the training stability. Second, the original MF fixes the classifier-free guidance scale during training, which sacrifices flexibility. We tackle this issue by formulating guidance as explicit conditioning variables, thereby retaining flexibility at test time. The diverse conditions are processed through in-context conditioning, which reduces model size and benefits performance. Overall, our improved MeanFlow (iMF) method, trained entirely from scratch, achieves 1.72 FID with a single function evaluation (1-NFE) on ImageNet 256x256. iMF substantially outperforms prior methods of this kind and closes the gap with multi-step methods while using no distillation. We hope our work will further advance fastforward generative modeling as a stand-alone paradigm.