LogCD：面向少步图像生成的局部到全局一致性蒸馏
LogCD: Local-to-global Consistency Distillation for Few-step Image Generation
两阶段局部到全局一致性蒸馏把 SDXL 这类潜扩散/整流流模型压到 2–4 步采样，且单模型可在步数间无缝切换，可直接复用到任何需要少步高保真生成的视觉管线（如实时文生图、编辑前置生成器）。
原文依据 LogCD accelerates SDXL to achieve a 33.5 CLIP score with just 3 sampling steps
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论将潜扩散模型（LDM）/整流流模型（RFM）蒸馏为可快速采样的模型正受到极大关注。然而，现有方法大多要么需要大量训练资源，要么导致质量下降，尤其是在文本-图像对齐方面。为应对这些挑战，我们提出局部到全局一致性蒸馏（LogCD），通过两阶段蒸馏加速 LDM/RFM。LogCD 先执行局部一致性蒸馏，再执行全局一致性蒸馏，以确保沿推理路径的一致性。此外，利用潜变量学习感知图像块相似度模型来增强感知一致性。值得注意的是，LogCD 具有高度灵活性，允许单个统一模型在 2 至 4 个采样步数下运行。在该范围内，模型性能随步数增加而平滑提升。仅用 70 个 A100 GPU 小时，LogCD 即加速 SDXL 在仅 3 个采样步数下达到 33.5 的 CLIP 分数，超越了使用更多步数的最优加速模型。
英文原文摘要 1093 字符
Distilling latent diffusion models (LDMs)/rectified flow models (RFMs) into ones that are fast to sample from conditions is attracting huge interest. However, the majority of existing methods either need significant training resources or lead to quality degradation, especially in text-image alignment. To address these challenges, we propose Local-to-global Consistency Distillation (LogCD) to accelerate LDMs/RFMs via two-stage distillation. LogCD first performs local consistency distillation and then executes global consistency distillation to ensure the consistency along inference path. Besides, Latent Learned Perceptual Image Patch Similarity model is exploited to enhance perceptual consistency. Notably, LogCD exhibits high flexibility, allowing a single unified model to operate with 2 to 4 sampling steps. The model's performance improves seamlessly as the number of steps increases within this range. With only 70 A100 GPU hours, LogCD accelerates SDXL to achieve a 33.5 CLIP score with just 3 sampling steps, surpassing state-of-the-art accelerated models using even more steps.