arXiv:2511.13387v3 Announce Type: replace 
Abstract: Denoising diffusion models have emerged as a dominant paradigm in image generation. Discretizing image data into tokens is a critical step for effectively integrating images with Transformer and other architectures. Although the Denoising Diffusion Codebook Models (DDCM) pioneered the use of pre-trained diffusion models for image tokenization, it strictly relies on the traditional discrete-time DDPM architecture. Consequently, it fails to adapt to modern continuous-time variants-such as Flow Matching and Consistency Models-and suffers from inefficient sampling in high-noise regions. To address these limitations, this paper proposes the Generalized Denoising Diffusion Codebook Models (gDDCM). We establish a unified theoretical framework and introduce a generic "De-noise and Back-trace" sampling strategy. By integrating a deterministic ODE denoising step with a residual-aligned noise injection step, our method resolves the challenge of adaptation. Furthermore, we introduce a backtracking parameter $p$ and significantly enhance tokenization ability. Extensive experiments on CIFAR10 and LSUN Bedroom datasets demonstrate that gDDCM achieves comprehensive compatibility with mainstream diffusion variants and significantly outperforms DDCM in terms of reconstruction quality and perceptual fidelity.

يمثل تقديم نماذج كود الكتابة للتصفية العامة (gDDCM) تقدمًا كبيرًا في تحويل الصور إلى رموز باستخدام نموذج تصفية مدرب مسبقًا. تعالج هذه الطريقة الجديدة قيود نماذج كود الكتابة للتصفية التقليدية (DDCM) من خلال دمج إطار نظري موحد واستراتيجية أخذ عينات جديدة تعزز الكفاءة في المناطق ذات الضوضاء العالية.

La introducción de los Modelos de Código de Difusión de Denoising Generalizados (gDDCM) marca un avance significativo en la tokenización de imágenes mediante el uso de un modelo de difusión preentrenado. Este nuevo enfoque aborda las limitaciones de los Modelos de Código de Difusión de Denoising (DDCM) tradicionales al integrar un marco teórico unificado y una nueva estrategia de muestreo que mejora la eficiencia en regiones de alto ruido.

L'introduction des Modèles de Codebook de Diffusion de Dénoyage Généralisés (gDDCM) représente une avancée significative dans la tokenisation d'images en utilisant un modèle de diffusion pré-entraîné. Cette nouvelle approche répond aux limitations des Modèles de Codebook de Diffusion de Dénoyage (DDCM) traditionnels en intégrant un cadre théorique unifié et une nouvelle stratégie d'échantillonnage qui améliore l'efficacité dans les régions à fort bruit.

The introduction of Generalized Denoising Diffusion Codebook Models (gDDCM) marks a significant advancement in image tokenization by utilizing a pre-trained diffusion model. This new approach addresses the limitations of traditional Denoising Diffusion Codebook Models (DDCM) by integrating a unified theoretical framework and a novel sampling strategy that enhances efficiency in high-noise regions.

Generalized Denoising Diffusion Codebook Models (gDDCM): Tokenizing images using a pre-trained diffusion model

Was this article worth reading? Share it

LucidQuery AI

Pixel Dojo

DataCircle.io

Dyad

Blunge

Deptho.ai

Ready to build your own newsroom?