TVM让生成模型一步到位，用新方法提升图像质量。
Terminal Velocity Matching
- 通过终端速度匹配建模任意两步间过渡，优化生成路径。
- 单次评估达3.29 FID，4次评估达1.99 FID，超越现有方法。
- 适合追求高效高质生成的科研与工程人员使用。
我们提出终端速度匹配（TVM），一种流匹配的泛化方法，实现高质量的一步和少步生成建模。TVM建模任意两个扩散时间步之间的过渡，并在终末时刻而非初始时刻对模型行为进行正则化。我们证明，当模型为Lipschitz连续时，TVM可提供数据分布与模型分布间$2$-Wasserstein距离的上界。然而，由于扩散Transformer不具备该性质，我们引入最小的结构修改，实现稳定且单阶段训练。为提升实用性，我们开发了融合注意力核，支持雅可比-向量积的反向传播，可良好扩展至Transformer架构。在ImageNet-256x256上，TVM以单次函数评估（NFE）取得3.29 FID，4次评估下达1.99 FID；在ImageNet-512x512上分别达4.32和2.94 FID，创下从零开始训练的一/少步模型最佳表现。
原文摘要 · Abstract (English)
We propose Terminal Velocity Matching (TVM), a generalization of flow matching that enables high-fidelity one- and few-step generative modeling. TVM models the transition between any two diffusion timesteps and regularizes its behavior at its terminal time rather than at the initial time. We prove that TVM provides an upper bound on the $2$-Wasserstein distance between data and model distributions when the model is Lipschitz continuous. However, since Diffusion Transformers lack this property, we introduce minimal architectural changes that achieve stable, single-stage training. To make TVM efficient in practice, we develop a fused attention kernel that supports backward passes on Jacobian-Vector Products, which scale well with transformer architectures. On ImageNet-256x256, TVM achieves 3.29 FID with a single function evaluation (NFE) and 1.99 FID with 4 NFEs. It similarly achieves 4.32 1-NFE FID and 2.94 4-NFE FID on ImageNet-512x512, representing state-of-the-art performance for one/few-step models from scratch.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。