改进流模型采样速度与质量，9次计算即可生成高质图像。
Simple ReFlow: Improved Techniques for Fast Flow Models
- 通过优化训练与推理流程，提升流模型采样效率
- 在仅9次函数求值下，各数据集FID达2.23~3.49
- 适合对生成速度与质量有严苛要求的应用场景
扩散模型和流匹配模型虽生成效果优异，但需大量采样步骤，限制了其在时间敏感任务中的应用。ReFlow可通过拉直生成轨迹加速采样，但其为迭代过程，通常需在模拟数据上训练，且易导致样本质量下降。本文系统分析ReFlow的设计空间，揭示以往启发式做法的潜在问题，并提出七项改进策略，涵盖训练动态、学习与推理优化。在CIFAR10（32×32）、AFHQv2（64×64）、FFHQ（64×64）及ImageNet-64上通过充分消融实验验证。结合全部技术，实现基于神经微分方程的快速生成最优表现：CIFAR10（无/有引导）FID为2.23 / 1.98，AFHQv2为2.30 / 1.91，FFHQ为2.84 / 2.67，ImageNet-64为3.49 / 1.74，均仅需9次神经函数评估。
原文摘要 · Abstract (English)
Diffusion and flow-matching models achieve remarkable generative performance but at the cost of many sampling steps, this slows inference and limits applicability to time-critical tasks. The ReFlow procedure can accelerate sampling by straightening generation trajectories. However, ReFlow is an iterative procedure, typically requiring training on simulated data, and results in reduced sample quality. To mitigate sample deterioration, we examine the design space of ReFlow and highlight potential pitfalls in prior heuristic practices. We then propose seven improvements for training dynamics, learning and inference, which are verified with thorough ablation studies on CIFAR10 $32 \times 32$, AFHQv2 $64 \times 64$, and FFHQ $64 \times 64$. Combining all our techniques, we achieve state-of-the-art FID scores (without / with guidance, resp.) for fast generation via neural ODEs: $2.23$ / $1.98$ on CIFAR10, $2.30$ / $1.91$ on AFHQv2, $2.84$ / $2.67$ on FFHQ, and $3.49$ / $1.74$ on ImageNet-64, all with merely $9$ neural function evaluations.
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。