Search papers, labs, and topics across Lattice.
This paper introduces Riemannian Residual Line Search, a novel method for one-step image editing that addresses the conflicting demands of aggressive and conservative transport updates in diffusion models. By treating the editing challenge as a candidate-selection problem, the authors enhance the edit quality through local time curvature estimation and CLIP alignment maximization. The method achieves state-of-the-art performance on the PIE-Bench++ evaluation, outperforming existing one-step update algorithms across various edit types.
Achieving state-of-the-art one-step image edits by cleverly balancing aggressive updates with conservative preservation techniques could redefine the capabilities of diffusion models.
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.
