DMGD: Train-Free Dataset Distillation with Semantic-Distribution Matching in Diffusion Models

2026-05-05

AI Summary

This work addresses the limitations of existing diffusion-based dataset distillation methods, which typically rely on additional fine-tuning and lack efficient guidance mechanisms. The authors propose a Dual-Matching Guided Diffusion (DMGD) framework that, for the first time, enables fully training-free dataset distillation within diffusion models. DMGD integrates semantic matching with optimal transport–based distribution matching, augmented by dynamic guidance and efficient approximation strategies to simultaneously preserve synthetic data diversity and enhance semantic alignment. Experimental results demonstrate that DMGD outperforms current state-of-the-art methods requiring fine-tuning by 2.1%, 5.4%, and 2.4% in average accuracy on ImageNet-Woof, ImageNet-Nette, and ImageNet-1K, respectively.

Abstract

Dataset distillation enables efficient training by distilling the information of large-scale datasets into significantly smaller synthetic datasets. Diffusion based paradigms have emerged in recent years, offering novel perspectives for dataset distillation. However, they typically necessitate additional fine-tuning stages, and effective guidance mechanisms remain underexplored. To address these limitations, we rethink diffusion based dataset distillation and propose a Dual Matching Guided Diffusion (DMGD) framework, centered on efficient training-free guidance. We first establish Semantic Matching via conditional likelihood optimization, eliminating the need for auxiliary classifiers. Furthermore, we propose a dynamic guidance mechanism that enhances the diversity of synthetic data while maintaining semantic alignment. Simultaneously, we introduce an optimal transport (OT) based Distribution Matching approach to further align with the target distribution structure. To ensure efficiency, we develop two enhanced strategies for diffusion based framework: Distribution Approximate Matching and Greedy Progressive Matching. These strategies enable effective distribution matching guidance with minimal computational overhead. Experimental results on ImageNet-Woof, ImageNet-Nette, and ImageNet-1K demonstrate that our training-free approach achieves significant improvements, outperforming state-of-the-art (SOTA) methods requiring additional fine-tuning by average accuracy gains of 2.1%, 5.4%, and 2.4%, respectively.

Problem

• dataset distillation
• diffusion models
• training-free
• semantic matching
• distribution matching

Innovation

• Dataset Distillation
• Diffusion Models
• Training-Free
• Semantic Matching
• Optimal Transport