Denoising Diffusion Bridge Models

Link
Abstract

Diffusion models are powerful generative models that map noise to data using stochastic processes. However, for many applications such as image editing, the model input comes from a distribution that is not random noise. As such, diffusion models must rely on cumbersome methods like guidance or projected sampling to incorporate this information in the generative process. In our work, we propose Denoising Diffusion Bridge Models (DDBMs), a natural alternative to this paradigm based on diffusion bridges, a family of processes that interpolate between two paired distributions given as endpoints. Our method learns the score of the diffusion bridge from data and maps from one endpoint distribution to the other by solving a (stochastic) differential equation based on the learned score. Our method naturally unifies several classes of generative models, such as score-based diffusion models and OT-Flow-Matching, allowing us to adapt existing design and architectural choices to our more general problem. Empirically, we apply DDBMs to challenging image datasets in both pixel and latent space. On standard image translation problems, DDBMs achieve significant improvement over baseline methods, and, when we reduce the problem to image generation by setting the source distribution to random noise, DDBMs achieve comparable FID scores to state-of-the-art methods despite being built for a more general task.

Synth

Problem:: 기존 확산 모델은 입력이 Random Noise로 고정되어 Image-to-Image Translation 같은 작업에 부적합함 / 기존 해결책들은 복잡하고 이론적으로 정립되지 않음

Solution:: 두 데이터 분포를 직접 잇는 Diffusion Bridge를 역방향으로 되돌리는 과정을 학습하는 DDBM(Denoising Diffusion Bridge Models) 프레임워크를 제안함 / 계산 가능한 Bridge의 Conditional Score를 매칭하는 Denoising Bridge Score Matching을 통해 모델을 학습시킴

Novelty:: 두 Endpoint의 통계량을 모두 고려하는 일반화된 파라미터화(Generalized Parameterization) 및 Scaling Function을 유도함 / 결정론적 ODE와 확률적 SDE 샘플링을 결합한 **하이브리드 샘플러(Hybrid Sampler)**를 제안하여 생성 품질을 향상시킴 / 기존 확산 모델과 Flow 기반 모델을 통합하는 일반적인 분포 변환 프레임워크를 제시함

Note:: Unconditional Generation은 Noise → Image니까 Image → Image 태스크를 위한 방법론을 제안해보자!

Summary

Motivation

Method

file-20250701180412467.png|650

Method 검증