PulseAugur
实时 03:35:56

新方法修复了扩散模型中无分类器引导的不稳定性

研究人员发现扩散模型中无分类器引导(CFG)存在一个关键问题,即高引导水平会导致过饱和和不稳定性。他们提出了一种新颖的修复机制,用修改后的版本替换标准的CFG公式,在不增加计算成本的情况下有效稳定了该过程。这种新方法在测试网格上相对于传统CFG取得了显著的改进,获得了9/9个FID点数胜利,并在稳定Stable Diffusion 1.5等模型的强引导场景方面显示出潜力。 AI

影响 这项研究提供了一种在不增加额外计算成本的情况下,提高扩散模型生成图像的稳定性和质量的方法。

排序理由 该集群包含一篇学术论文,详细介绍了一种改进现有AI模型的新方法。

在 arXiv cs.LG 阅读 →

AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →

新方法修复了扩散模型中无分类器引导的不稳定性

报道来源 [3]

  1. arXiv cs.LG TIER_1 English(EN) · Shiheng Zhang ·

    Guidance Breaks the Fitted Operator: A Terminal-Fitted Repair for Classifier-Free Guidance

    arXiv:2607.07665v1 Announce Type: new Abstract: Classifier-free guidance (CFG) is the standard way to strengthen class-conditioning in diffusion and flow-matching samplers, yet at large guidance it oversaturates and destabilizes, symptoms practitioners suppress with more steps or…

  2. arXiv cs.LG TIER_1 English(EN) · Shiheng Zhang ·

    Guidance Breaks the Fitted Operator: A Terminal-Fitted Repair for Classifier-Free Guidance

    Classifier-free guidance (CFG) is the standard way to strengthen class-conditioning in diffusion and flow-matching samplers, yet at large guidance it oversaturates and destabilizes, symptoms practitioners suppress with more steps or limited-interval schedules. We analyze CFG thro…

  3. Hugging Face Daily Papers TIER_1 English(EN) ·

    Guidance Breaks the Fitted Operator: A Terminal-Fitted Repair for Classifier-Free Guidance

    Classifier-free guidance (CFG) is the standard way to strengthen class-conditioning in diffusion and flow-matching samplers, yet at large guidance it oversaturates and destabilizes, symptoms practitioners suppress with more steps or limited-interval schedules. We analyze CFG thro…