通过对齐扩散逆链实现真实世界图像变体生成
Real-World Image Variation by Aligning Diffusion Inversion Chain
AI总结：
针对扩散模型生成图像与真实图像存在域差距导致真实图像变体生成质量不佳的问题，提出RIVAL推理流水线，通过跨图像自注意力注入和逐步分布归一化对齐扩散逆链，无需额外优化即可生成高质量变体，性能优于现有方法且可推广到其他扩散生成任务。
摘要：
近期扩散模型的发展已实现借助文本提示生成高保真图像。然而，生成图像与真实世界图像之间存在域差距，这给生成高质量的真实世界图像变体带来了挑战。我们的研究发现，这种域差距源于不同扩散过程中的潜变量分布差距。为解决这一问题，我们提出了一种名为Real-world Image Variation by ALignment（RIVAL，基于对齐的真实世界图像变体生成）的新型推理流水线，它利用扩散模型从单张图像样本生成图像变体。我们的流水线通过将图像生成过程与源图像的逆链对齐，提升图像变体的生成质量。具体而言，我们证明了逐步潜变量分布对齐对于生成高质量变体至关重要。为实现这一点，我们设计了用于特征交互的跨图像自注意力注入，以及用于对齐潜变量特征的逐步分布归一化。将这些对齐过程整合到扩散模型中，使得RIVAL无需额外参数优化即可生成高质量图像变体。我们的实验结果表明，所提出的方法在语义相似度和感知质量方面优于现有方法。这种通用的推理流水线可轻松应用于其他基于扩散的生成任务，例如图像条件的文本到图像生成和风格化。
英文摘要：
Recent diffusion model advancements have enabled high-fidelity images to be generated using text prompts. However, a domain gap exists between generated images and real-world images, which poses a challenge in generating high-quality variations of real-world images. Our investigation uncovers that this domain gap originates from a latents' distribution gap in different diffusion processes. To address this issue, we propose a novel inference pipeline called Real-world Image Variation by ALignment (RIVAL) that utilizes diffusion models to generate image variations from a single image exemplar. Our pipeline enhances the generation quality of image variations by aligning the image generation process to the source image's inversion chain. Specifically, we demonstrate that step-wise latent distribution alignment is essential for generating high-quality variations. To attain this, we design a cross-image self-attention injection for feature interaction and a step-wise distribution normalization to align the latent features. Incorporating these alignment processes into a diffusion model allows RIVAL to generate high-quality image variations without further parameter optimization. Our experimental results demonstrate that our proposed approach outperforms existing methods concerning semantic similarity and perceptual quality. This generalized inference pipeline can be easily applied to other diffusion-based generation tasks, such as image-conditioned text-to-image generation and stylization.
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。
