arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2606.24844cs.CV

弥合流形差距:黎曼残差线搜索用于一步图像编辑

Bridging the Manifold Gap: Riemannian Residual Line Search for One-Step Image Editing

UCAS(中国科学院大学);WashU(华盛顿大学);SHU(上海大学)

Hongzhu Yi, Zhongtian Luo, Tong Li, Yiyan Fan, Jungang Xu

AI总结:

针对一步扩散编辑中固定更新强度无法兼顾目标提示和源图像保真度的问题,提出黎曼残差线搜索方法,通过局部时间曲率估计和残差路径选择,在PIE-Bench++上达到SOTA。

摘要:

一步扩散编辑器速度快,因为它们避免了反演和迭代优化,但单次传输更新必须足够激进以实现目标提示,同时足够保守以保留源图像——而固定的更新强度无法满足所有编辑类型的需求。我们将这种张力视为能量场传输之上的事后候选选择问题,而不是一个新的编辑模型。我们提出的方法,黎曼残差线搜索,首先通过估计提示-增量场的局部时间曲率,并将校正后的方向投影回原始一阶能量场传输估计的更新范数上,构建一个更强的编辑。然后,它从源图像到该强编辑形成一条小的残差路径,保留原始一阶输出作为候选之一,并通过最大化目标提示的CLIP对齐来选择最终图像。在包含10种编辑类型ID的700样本PIE-Bench++评估中,我们的方法在当前一步更新算法中达到了最先进的性能。

英文摘要:

One-step diffusion editors are fast because they avoid inversion and iterative optimization, but a single transport update must be aggressive enough to realize the target prompt and conservative enough to preserve the source image--and no fixed update strength satisfies both demands across edit types. We treat this tension as a post-hoc candidate-selection problem on top of energy-field transport rather than as a new editing model. Our proposed method, Riemannian Residual Line Search, first builds a stronger edit by estimating the local time curvature of the prompt-delta field and projecting the corrected direction back onto the update norm of the original first-order energy-field transport estimation. It then forms a small residual path from the source image to this strong edit, retains the original first-order output as one candidate, and picks the final image by maximizing target-prompt CLIP alignment. On a 700-sample PIE-Bench++ evaluation across 10 edit type IDs, our method achieves state-of-the-art (SOTA) performance among current one-step update algorithms.

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。

↑