发表机构
DGIST（大邱庆北科学技术院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究真实图像扩散反演的质量-成本权衡，通过高斯噪声锚分离出压缩不对称与轨迹绑定机制，据此提出NARC方法，该方法无需训练，存储少且效果好，在实验中表现出色并可应用于其他模型。
AI 中文摘要
真实图像扩散反演受质量-成本权衡的严格制约，涉及计算、存储或逐图像优化等成本。我们通过定义扩散轨迹的前向高斯噪声锚来研究这种权衡，并分离出有效存储噪声反演背后的两种机制。首先，扩散噪声呈现元素级压缩不对称：int8全维锚可保留重建，而低维子空间摘要可靠性低得多，甚至在相当或更小有效载荷时也常崩溃；这种元素级优于子空间排序在五种存储噪声反演方法中都存在。其次，反演是轨迹绑定且与得分先验耦合的：匹配前向锚和训练得分网络都必要，反对纯代数恒等式解释。这些发现明确了存储内容及使用方式，从而产生了噪声锚定反向校正（NARC），一种无需训练的反演原语，存储单个int8潜在锚并按固定的、依赖噪声水平的锚权重调度重复使用：反向轨迹以噪声为主时强锚定，图像细节出现时放松锚定。在使用Stable Diffusion 1.5的PIE - Bench++上，NARC优于五种现代非精确基线，比PnP DirectInv提高PSNR 3.24 dB，同时反演存储比PnP DirectInv少约400倍。压缩不对称性、锚特异性和编辑插件也适用于SDXL 1024^2。
英文摘要
Real-image diffusion inversion is governed by a tight quality-cost trade-off, with costs incurred in computation, storage, or per-image optimization. We study this trade-off through the forward Gaussian noise anchor that defines a diffusion trajectory and isolate two mechanisms behind effective stored-noise inversion. First, diffusion noise exhibits an element-wise compression asymmetry: int8 full-dimensional anchors preserve reconstruction, whereas low-dimensional subspace summaries are much less reliable, often collapsing even at comparable or smaller payloads; the element-wise over subspace ordering persists across five stored-noise inversion methods. Second, inversion is trajectory-bound and score-prior coupled: the matched forward anchor and a trained score network are both necessary, arguing against a purely algebraic-identity explanation. Together, these findings specify what to store and how to use it. They lead to Noise-Anchored Reverse Correction (NARC), a training-free inversion primitive that stores a single int8 latent anchor and reuses it with a fixed, noise-level-dependent anchor-weight schedule: strong anchoring when the reverse trajectory is noise-dominated, then relaxed anchoring as image detail emerges. On PIE-Bench++ with Stable Diffusion 1.5, NARC outperforms five modern non-exact baselines and improves PSNR by +3.24 dB over PnP DirectInv while using about 400x less inversion storage than PnP DirectInv. The compression asymmetry, anchor specificity, and editing plug-in also transfer to SDXL 1024^2.
