空间自适应噪声注入
Spatially Adaptive Noise Injection
- University of Applied Sciences Western Switzerland, HES-SO Geneva（瑞士西部应用科学与艺术大学日内瓦校区）
- University of Geneva（日内瓦大学）
机构由 AI 辅助整理，请以论文原文为准。
AI总结：
针对扩散采样中均匀噪声注入忽视图像几何的问题，提出空间自适应噪声注入（SANI）框架，通过概率门控与逐像素自适应方差动态调整噪声，提升FID并保持竞争力。
AI中文摘要：
扩散采样器通过随机（DDPM）或确定性（DDIM）更新来逆转学习到的加噪过程，这两种更新方式代表了由标量噪声注入方差控制的单一族的两端，该方差在每个空间位置以相同方式应用。这种统一方法忽略了自然图像的几何特性：高曲率区域（如边缘和纹理）中，去噪器不确定，受益于随机校正；而平滑区域中，分数精确，注入噪声反而会降低质量。本研究探讨了在给定时间步中每个像素是否需要随机校正，并提出了空间自适应噪声注入（SANI），一种新颖的采样框架，能够动态调整逐像素的噪声应用。SANI集成了概率门控机制与推导出的空间自适应方差，确保在需要的地方精确注入噪声以细化复杂特征，同时保留结构良好的区域。实验结果和解耦消融研究表明，SANI在不同采样时间步上始终优于原始DDPM和DDIM端点采样器的弗雷歇初始距离（FID），并与方差学习基线保持竞争力，凸显了扩散采样中空间自适应性的重要性。
英文摘要：
Diffusion samplers reverse a learned noising process using either stochastic (DDPM) or deterministic (DDIM) updates, which represent endpoints of a single family controlled by a scalar noise-injection variance that is applied identically at every spatial location. This uniform approach neglects the geometry of natural images: high-curvature regions such as edges and textures, where the denoiser is uncertain, benefit from stochastic correction, whereas smooth regions, where the score is precise, are degraded by injected noise. This work investigates whether each pixel requires stochastic correction at a given timestep and introduces Spatially Adaptive Noise Injection (SANI), a novel sampling framework that dynamically adjusts noise application on a per-pixel basis. SANI integrates a probabilistic gating mechanism with a derived spatially adaptive variance, ensuring that noise is injected precisely where needed to refine complex features while preserving well-formed structures. Experimental results and decoupling ablations demonstrate that SANI consistently improves Fréchet Inception Distance (FID) over the vanilla DDPM and DDIM endpoint samplers across diverse sampling timesteps, while remaining competitive with variance-learning baselines, highlighting the importance of spatial adaptivity in diffusion sampling.
