arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2301.10972cs.CVcs.GRcs.LGcs.MM
收藏

论噪声调度对扩散模型的重要性

On the Importance of Noise Scheduling for Diffusion Models

Google Research(谷歌研究院)

Ting Chen

AI总结:

本文实证研究噪声调度对扩散模型的影响,发现最优调度随图像尺寸变化,提出通过缩放输入数据实现跨尺寸泛化,结合 RIN 在 ImageNet 上实现 1024×1024 单阶段端到端生成。

摘要:

我们实证研究了噪声调度策略对去噪扩散生成模型的影响。得出三个发现:(1)噪声调度对性能至关重要,且最优调度取决于任务(如图像尺寸);(2)当增大图像尺寸时,最优噪声调度会向更嘈杂的方向偏移(由于像素冗余增加);(3)在保持噪声调度函数固定的情况下,将输入数据简单缩放 $b$ 倍(相当于将 logSNR 偏移 $\log b$)是一种跨图像尺寸的良好策略。这一简单方案与近期提出的 Recurrent Interface Network (RIN) 结合,在 ImageNet 上实现了基于像素的最先进扩散模型,支持在 1024×1024 分辨率下进行单阶段、端到端的多样高保真图像生成(无需上采样/级联)。

英文摘要:

We empirically study the effect of noise scheduling strategies for denoising diffusion generative models. There are three findings: (1) the noise scheduling is crucial for the performance, and the optimal one depends on the task (e.g., image sizes), (2) when increasing the image size, the optimal noise scheduling shifts towards a noisier one (due to increased redundancy in pixels), and (3) simply scaling the input data by a factor of $b$ while keeping the noise schedule function fixed (equivalent to shifting the logSNR by $\log b$) is a good strategy across image sizes. This simple recipe, when combined with recently proposed Recurrent Interface Network (RIN), yields state-of-the-art pixel-based diffusion models for high-resolution images on ImageNet, enabling single-stage, end-to-end generation of diverse and high-fidelity images at 1024$\times$1024 resolution (without upsampling/cascades).

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑