SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations

Link
Abstract

Guided image synthesis enables everyday users to create and edit photo-realistic images with minimum effort. The key challenge is balancing faithfulness to the user inputs (e.g., hand-drawn colored strokes) and realism of the synthesized images. Existing GAN-based methods attempt to achieve such balance using either conditional GANs or GAN inversions, which are challenging and often require additional training data or loss functions for individual applications. To address these issues, we introduce a new image synthesis and editing method, Stochastic Differential Editing (SDEdit), based on a diffusion model generative prior, which synthesizes realistic images by iteratively denoising through a stochastic differential equation (SDE). Given an input image with user guide in a form of manipulating RGB pixels, SDEdit first adds noise to the input, then subsequently denoises the resulting image through the SDE prior to increase its realism. SDEdit does not require task-specific training or inversions and can naturally achieve the balance between realism and faithfulness. SDEdit outperforms state-of-the-art GAN-based methods by up to 98.09% on realism and 91.72% on overall satisfaction scores, according to a human perception study, on multiple tasks, including stroke-based image synthesis and editing as well as image compositing.

Synth

Problem:: 기존 Guided Image Synthesis 방법들은 사용자의 입력에 대한 Faithfulness와 생성된 이미지의 Realism 사이의 균형을 맞추기 어려움

Solution:: 사용자 가이드 이미지에 노이즈가 추가된 이미지를 초기값으로 사용하여 t0 시점부터 Reverse SDE를 실행하여 이미지를 Denoising

Novelty:: 별도의 Task-Specific 학습 데이터나 손실 함수 없이 다양한 편집 작업 수행 가능 / 완전한 노이즈가 아닌, 노이즈가 추가된 사용자 가이드로부터 Reverse SDE를 시작하는 방식 제안 / Reverse SDE 시작 시점(t0) 조절을 통해 Realism과 Faithfulness 간의 균형을 자연스럽게 제어함

Note:: Editing의 기본이 되는 연구, 방법이 간단하고 직관적임

Summary

Motivation

Method

배경: SDE 기반 이미지 생성 (기존 방식)

SDEdit: SDE 기반 이미지 편집 (제안 방식)

file-20250425041644623.png|775

가이드에 노이즈를 추가해서 이미지와 Stroke를 가깝게 만듦 → 가까워 졌으니 Denoising에서 Stroke보다 Image에 가깝게 이동시킴

Method 검증