Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Bahram, Yara, Desbos, Mélodie, Shateri, Mohammadhadi, Granger, Eric
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2511.18281
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910071663886336
author Bahram, Yara
Desbos, Mélodie
Shateri, Mohammadhadi
Granger, Eric
author_facet Bahram, Yara
Desbos, Mélodie
Shateri, Mohammadhadi
Granger, Eric
contents Diffusion models (DMs) produce high-quality images, yet their sampling remains costly when adapted to new domains. Distilled DMs are faster but typically remain confined within their teacher's domain. Thus, fast and high-quality generation for novel domains relies on two-stage pipelines: Adapt-then-Distill or Distill-then-Adapt. However, both add design complexity and often degrade quality or diversity. We introduce Uni-DAD, a single-stage pipeline that unifies DM distillation and adaptation. It couples two training signals: (i) a dual-domain distribution-matching distillation (DMD) objective that guides the student toward the distributions of the source teacher and a target teacher, and (ii) a multi-head generative adversarial network (GAN) loss that encourages target realism across multiple feature scales. The source domain distillation preserves diverse source knowledge, while the multi-head GAN stabilizes training and reduces overfitting, especially in few-shot regimes. The inclusion of a target teacher facilitates adaptation to more structurally distant domains. We evaluate Uni-DAD on two comprehensive benchmarks for few-shot image generation (FSIG) and subject-driven personalization (SDP) using diffusion backbones. It delivers better or comparable quality to state-of-the-art (SoTA) adaptation methods even with less than 4 sampling steps, and often surpasses two-stage pipelines in quality and diversity. Code: https://github.com/yaramohamadi/uni-DAD.
format Preprint
id arxiv_https___arxiv_org_abs_2511_18281
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Uni-DAD: Unified Distillation and Adaptation of Diffusion Models for Few-step Few-shot Image Generation
Bahram, Yara
Desbos, Mélodie
Shateri, Mohammadhadi
Granger, Eric
Computer Vision and Pattern Recognition
Artificial Intelligence
Diffusion models (DMs) produce high-quality images, yet their sampling remains costly when adapted to new domains. Distilled DMs are faster but typically remain confined within their teacher's domain. Thus, fast and high-quality generation for novel domains relies on two-stage pipelines: Adapt-then-Distill or Distill-then-Adapt. However, both add design complexity and often degrade quality or diversity. We introduce Uni-DAD, a single-stage pipeline that unifies DM distillation and adaptation. It couples two training signals: (i) a dual-domain distribution-matching distillation (DMD) objective that guides the student toward the distributions of the source teacher and a target teacher, and (ii) a multi-head generative adversarial network (GAN) loss that encourages target realism across multiple feature scales. The source domain distillation preserves diverse source knowledge, while the multi-head GAN stabilizes training and reduces overfitting, especially in few-shot regimes. The inclusion of a target teacher facilitates adaptation to more structurally distant domains. We evaluate Uni-DAD on two comprehensive benchmarks for few-shot image generation (FSIG) and subject-driven personalization (SDP) using diffusion backbones. It delivers better or comparable quality to state-of-the-art (SoTA) adaptation methods even with less than 4 sampling steps, and often surpasses two-stage pipelines in quality and diversity. Code: https://github.com/yaramohamadi/uni-DAD.
title Uni-DAD: Unified Distillation and Adaptation of Diffusion Models for Few-step Few-shot Image Generation
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2511.18281