Increasing the Utility of Synthetic Images through Chamfer Guidance

2025-08-14

AI Summary

To address the limitations of synthetic images as training data—namely, low quality, insufficient diversity, and distributional shift from real data—this paper proposes Chamfer Guidance: a training-free, few-shot generation guidance mechanism grounded in the Chamfer distance. Leveraging only two real images, it computes geometric discrepancies between synthetic and real distributions via nearest-neighbor matching and explicitly steers conditional generation. Compared to unconditional synthesis, our method significantly improves both sample diversity and distribution alignment, eliminates the need for auxiliary model training, and reduces sampling FLOPs by 31%. On ImageNet-1K, using merely two real images, it achieves 96.4% classification accuracy and 86.4% distribution coverage. Downstream task accuracy improves by up to 15% (in-distribution) and 16% (out-of-distribution). To our knowledge, this is the first approach enabling efficient, lightweight, and high-fidelity few-shot synthetic data generation.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Conditional image generative models hold considerable promise to produce infinite amounts of synthetic training data. Yet, recent progress in generation quality has come at the expense of generation diversity, limiting the utility of these models as a source of synthetic training data. Although guidance-based approaches have been introduced to improve the utility of generated data by focusing on quality or diversity, the (implicit or explicit) utility functions oftentimes disregard the potential distribution shift between synthetic and real data. In this work, we introduce Chamfer Guidance: a training-free guidance approach which leverages a handful of real exemplar images to characterize the quality and diversity of synthetic data. We show that by leveraging the proposed Chamfer Guidance, we can boost the diversity of the generations w.r.t. a dataset of real images while maintaining or improving the generation quality on ImageNet-1k and standard geo-diversity benchmarks. Our approach achieves state-of-the-art few-shot performance with as little as 2 exemplar real images, obtaining 96.4% in terms of precision, and 86.4% in terms of distributional coverage, which increase to 97.5% and 92.7%, respectively, when using 32 real images. We showcase the benefits of the Chamfer Guidance generation by training downstream image classifiers on synthetic data, achieving accuracy boost of up to 15% for in-distribution over the baselines, and up to 16% in out-of-distribution. Furthermore, our approach does not require using the unconditional model, and thus obtains a 31% reduction in FLOPs w.r.t. classifier-free-guidance-based approaches at sampling time.

Problem

• Balancing quality and diversity in synthetic image generation
• Addressing distribution shift between synthetic and real data
• Improving utility of synthetic data for training classifiers

Innovation

• Chamfer Guidance improves synthetic data utility
• Leverages real exemplars for quality and diversity
• Reduces FLOPs by 31% vs classifier-free guidance