arXiv cs.AI· Duc Vu, Kien Nguyen, Ngan Nguyen, Trong-Tung Nguyen, Phong Nguyen, Khoi Nguyen, Cuong Pham, Anh Tran·· 2 天前AI 评分26
InverFill：一步反演增强少步扩散模型图像修复
InverFill: One-Step Inversion for Enhanced Few-Step Diffusion Inpainting
AI 导读
InverFill 是一种面向图像修复的一步反演方法，通过将输入掩码图像的语义信息注入初始噪声，解决少步扩散模型因随机高斯噪声初始化导致的语义错位与保真度下降问题。该方法无需训练修复模型，仅以语义对齐噪声驱动少步文生图模型的混合采样流程，在低 NFE 下显著优于原始混合采样，甚至媲美专用修复模型。InverFill 无需真实图像监督，仅增加极少推理开销，已被 CVPR'26 主会议接收。
正文 · 原文
Abstract:Recent diffusion-based models achieve photorealism in image inpainting but require many sampling steps, limiting practical use. Few-step text-to-image models offer faster generation, but naively applying them to inpainting yields poor harmonization and artifacts between the background and inpainted region. We trace this cause to random Gaussian noise initialization, which under low function evaluations causes semantic misalignment and reduced fidelity. To overcome this, we propose InverFill, a one-step inversion method tailored for inpainting that injects semantic information from the input masked image into the initial noise, enabling high-fidelity few-step inpainting. Instead of training inpainting models, InverFill leverages few-step text-to-image models in a blended sampling pipeline with semantically aligned noise as input, significantly improving vanilla blended sampling and even matching specialized inpainting models at low NFEs. Moreover, InverFill does not require real-image supervision and only adds minimal inference overhead. Extensive experiments show that InverFill consistently boosts baseline few-step models, improving image quality and text coherence without costly retraining or heavy iterative optimization.
| Comments: | Accepted to CVPR'26 (Main Conference) |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2603.23463 [cs.CV] |
| (or arXiv:2603.23463v3 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2603.23463 arXiv-issued DOI via DataCite |
Submission history
From: Duc Vu [view email]
[v1]
Tue, 24 Mar 2026 17:32:55 UTC (43,237 KB)
[v2]
Fri, 21 Aug 2026 11:04:17 UTC (43,237 KB)
[v3]
Thu, 15 Oct 2026 17:30:56 UTC (43,237 KB)
来源：arXiv cs.AI · arxiv.org